Mixture-of-Experts (MoE) architecture with experts parallelism scales LLMs efficiently by activating only a subset of experts per input, avoiding proportional training costs. However, the intensive and heterogeneous communication substantially hinders the efficiency and scalability of MoE training in the resource-constrained scenario. Existing communication compression techniques fall short in MoE training due to: (i) Intensive training amplifies compression overhead, compromising training efficiency; (ii) Accumulated compression errors propagate through the network, degrading training quality. In this paper, we propose RCMoE, a communication-efficient Random Compression framework for MoE training with two core modules: (1) Local-Stochastic Quantization compresses the all-to-all communication by stochastically quantizing each row of the expert's intermediate computing results in parallel, effectively improving the compression efficiency and reducing compression error; (2) Probabilistic Thresholding Sparsification compresses the all-reduce communication by probabilistically sampling large gradients at high probability, thereby reducing the computational complexity and maintaining the convergence efficiency. Experiments on four typical MoE training tasks prove that RCMoE achieves higher 5.9x-8.1x total communication compression ratios and 1.3x-10.1x training speedup compared with the state-of-the-art compression techniques while maintaining the MoE training accuracy.

Handling communication overhead in large-scale tensor-parallel training remains a critical challenge due to the dense, near-zero distributions of intermediate tensors, which exacerbate errors under frequent communication and introduce significant computational overhead during compression. To this end, we propose TACO (Tensor-parallel Adaptive COmmunication compression), a robust FP8-based framework for compressing TP intermediate tensors. First, we employ a data-driven reshaping strategy combined with an Adaptive Scale–Hadamard Transform to enable high-fidelity FP8 quantization, while its Dual-Scale Quantization mechanism ensures numerical stability throughout training. Second, we design a highly fused compression operator to reduce memory traffic and kernel launch overhead, allowing efficient overlap with communication. Finally, we integrate TACO with existing state-of-the-art methods for Data and Pipeline Parallelism to develop a compression-enabled 3D-parallel training framework. Detailed experiments on GPT models and Qwen model demonstrate up to 1.87 × end-to-end throughput improvement while maintaining near-lossless accuracy, validating the effectiveness and efficiency of TACO in large-scale training.

AI-driven methods have demonstrated considerable success in tackling the central challenge of accurately solving the Schrödinger equation for complex many-body systems. Among neural network quantum state (NNQS) approaches, the NNQS-SCI (Selected Configuration Interaction) method stands out as a state-of-the-art technique, recognized for its high accuracy and scalability. However, its application to larger systems is severely constrained by a hybrid CPU-GPU architecture. Specifically, centralized CPU-based global de-duplication creates a severe scalability barrier due to communication bottlenecks, while host-resident coupled-configuration generation induces prohibitive computational overheads. We introduce QiankunNet-cuSCI, a fully GPU-accelerated SCI framework designed to overcome these bottlenecks. It first integrates a distributed, load-balanced global de-duplication algorithm to minimize redundancy and communication overhead at scale. To address compute limitations, it employs specialized, fine-grained CUDA kernels for exact coupled configuration generation. Finally, to break the single-GPU memory barrier exposed by this full acceleration, it incorporates a GPU memory-centric runtime featuring GPU-side pooling, streaming mini-batches, and overlapped offloading. This design enables much larger configuration spaces and shifts the bottleneck from host-side limitations back to on-device inference. Our evaluation demonstrates that our work fundamentally expands the scale of solvable problems. On an NVIDIA A100 cluster with 64 GPUs, our work achieves up to 2.32 × end-to-end speedup over the highly-optimized NNQS-SCI baseline while preserving the same chemical accuracy. Furthermore, it demonstrates excellent distributed performance, maintaining over 90% parallel efficiency in strong scaling tests.

With the exponential growth of computing power, large-scale scientific simulations are producing massive volumes of data, leading to critical storage and I/O challenges. Error-bounded lossy compression has become one of the most effective solutions for reducing data size while preserving accuracy. Meanwhile, to achieve high-performance compression on such large datasets, leveraging GPUs has become increasingly essential. GPU-based lossy compressors deliver strong performance, but typically support only single-precision decompression, limiting their ability to meet the diverse accuracy requirements of scientific workflows. Progressive compressors can address this limitation by enabling on-demand precision retrieval. However, existing progressive lossy compressors on GPU still suffer from low throughput. To overcome these challenges, we present PRISM, a GPU-based progressive lossy compressor that achieves both high throughput and multi-precision retrieval, which introduces a high performance progressive framework that integrates the multiple interpolation predictors, efficient bitplane extraction, and an enhanced lossless compression that combines sign-absolute coding with zero-aware parallel algorithms. Evaluations on representative real-world datasets from five scientific domains show that PRISM significantly outperforms state-of-the-art progressive compressors on GPU, reducing retrieval data volume by over 15.6× and achieving up to 20.1× higher throughput on the NVIDIA H100 GPU under the same error bounds.

As training scales grow, collective communication libraries (CCL) increasingly face anomalies arising from complex interactions among hardware, software, and environmental factors. These anomalies typically manifest as slow/hang communication, the most frequent and time-consuming category to diagnose. However, traditional diagnostic methods remain inaccurate and inefficient, frequently requiring hours or even days for root cause analysis. To address this, we propose CCL-D, a high-precision diagnostic system designed to detect and locate slow/hang anomalies in large-scale distributed training. CCL-D integrates a rank-level real-time probe with an intelligent decision analyzer. The probe measures cross-layer anomaly metrics using a lightweight distributed tracing framework to monitor communication traffic. The analyzer performs automated anomaly detection and root-cause location, precisely identifying the faulty GPU rank. Deployed on a 4,000-GPU cluster over one year, CCL-D achieved near-complete coverage of known slow/hang anomalies and pinpointed affected ranks within 6 minutes—substantially outperforming existing solutions.

Collective communication is critical to scaling large language model (LLM) training across various parallelism strategies, including data, tensor, and pipeline parallelism on GPU clusters. However, as model sizes and training scales increase, communication overhead is emerging as a major performance bottleneck. While compression is a promising mitigation strategy, existing solutions often lack user-transparency, hinder deployment and extensibility, and are not co-designed with communication algorithms. To address these limitations, we present COCCL, a high-performance collective communication library built on top of NCCL. COCCL introduces a novel programming model that can easily integrate compression into communication workflows with flexible configurability. It features a suite of compression-aware collective algorithms and runtime overlap mechanisms that mitigate error propagation and reduce computational overhead. We integrate well-established compression techniques into COCCL and tune the compression configurations during 3D-parallel training on GPT and Qwen models with up to 7 billion parameters. Using the optimal configuration (COCCL-3D), we achieve 1.24× throughput improvement while maintaining training accuracy.

Efficient CPU power capping is essential for energy saving and fault tolerance in parallel computing clusters, but its effectiveness depends on accurate and timely processor power forecasting with minimal sampling overhead. Existing methods often struggle to balance these factors under scalability constraints, as hardware limitations tightly bound the available sampling resources. This article focuses on the issue of high-precision real-time processor power forecasting while maintaining (or minimally increasing) the total overhead of multiprocessor power forecasting, particularly when the parallelism scale ranges from P to 2P processors or when the problem size scales from M to 2M . We propose CML-PowF , a low-overhead multiprocessor real-time power forecasting approach based on data clustering. CML-PowF integrates two key algorithms: Alg-CEF , which conducts cluster matching on the runtime characteristics of the program at the P / M scale, and models the tradeoff among forecasting error, time span, and sampling overhead. Alg-MSF , which leverages execution patterns from smaller-scale runs to determine the optimal sampling overhead and forecasting time span at the 2P / 2M scale. We evaluate CML-PowF on x86 and ARM platforms with up to 32 computing nodes (2,048 cores). Results show that it achieves 3–6% forecasting error at large scales with only 0.2–0.5% degradation compared to the P / M scale, without increasing total sampling overhead. Integrated with the PowC control system, CML-PowF effectively maintains real-time processor power below target thresholds.

Finding frequent subgraph patterns in a big graph is an important problem with many applications such as classifying chemical compounds and building indexes to speed up graph queries. Since this problem is NP-hard, some recent parallel and distributed systems have been developed to accelerate the mining. However, they often have a huge memory cost, very long running time, suboptimal load balancing, poor scale-out capability, and possibly inaccurate results. In this article, we propose an efficient system called T-FSM for parallel mining of frequent subgraph patterns in a big graph. T-FSM supports a new anti-monotonic frequentness measure called Fraction-Score, which is more accurate than the widely used MNI measure. The execution engine of T-FSM supports both intra-machine parallelism and inter-machine parallelism. For intra-machine parallelism, T-FSM adopts a novel task-based execution model to ensure high multithreading concurrency, bounded memory consumption, and effective load balancing. For inter-machine parallelism, T-FSM ensures good scale-out performance with a lightweight pattern rebalancing approach that reduces workload skewness of pattern evaluations among machines. To avoid recomputing the contexts for migrated patterns, we design a novel context cache table to support concurrent and asynchronous requesting and caching of remote context data, which can timely evict and garbage collect used pattern contexts that are no longer needed to keep memory consumption bounded. Extensive experiments show that T-FSM is orders of magnitude faster than existing state-of-the-art parallel systems (more than 10×, 51×, 131×, 55× speedup over ScaleMine, DistGraph, Pangolin and Peregrine, respectively) and distributed systems (more than 42× and 88× over ScaleMine and DistGraph, respectively) for frequent subgraph pattern mining, and it scales out satisfactorily to 512 CPU cores on the Polaris supercomputer at Argonne National Laboratory.

Burst buffers (BBs) act as an intermediate storage layer between compute nodes and parallel file systems (PFS), effectively alleviating the I/O performance gap in high-performance computing (HPC). As scientific simulations and AI workloads generate larger checkpoints and analysis outputs, BB capacity shortages and PFS bandwidth bottlenecks are emerging, and CPU-based compression is not an effective solution due to its high overhead. We introduceComputational Burst Buffers(CBBs), a storage paradigm that embeds hardware compression engines such as application-specific integrated circuit (ASIC) inside computational storage drives (CSDs) at the BB tier. CBB transparently offloads both lossless and error-bounded lossy compression from CPUs to CSDs, thereby (i) expanding effective SSD-backed BB capacity, (ii) reducing BB–PFS traffic, and (iii) eliminating contention and energy overheads of CPU-based compression. Unlike prior CSD-based compression designs targeting databases or flash caching, CBB co-designs the burst-buffer layer and CSD hardware for HPC and quantitatively evaluates compression offload in BB–PFS hierarchies. We prototype CBB using a PCIe 5.0 CSD with an ASIC Zstd-like compressor and an FPGA prototype of an SZ entropy encoder, and evaluate CBB on a 16-node cluster. Experiments with four representative HPC applications and a large-scale workflow simulator show up to 61% lower application runtime, 8–12× higher cache hit ratios, and substantially reduced compute-node CPU utilization compared to software compression and conventional BBs. These results demonstrate that compression-aware BBs with CSDs provide a practical, scalable path to next-generation HPC storage.

Compared to replication-based storage systems, erasure-coded storage incurs significantly higher overhead during data updates. To address this issue, various parity logging methods have been proposed. Nevertheless, due to the long update path and substantial amount of random I/O involved in erasure code update processes, the resulting long latency and low through put often fail to meet the requirements of high performance applications. To address this challenge, we propose TSUE+, an efficient update framework with a swift recycling mechanism. TSUE+ divides the update process into two distinct stages: in the synchronous stage, data updates are stored in the format of replica data logs, eliminating random I/O by trading space for time; in the asynchronous stage, the recorded update logs are recycled and merged into original data and parity blocks, thereby reclaiming the storage overhead incurred in the synchronization phase. By converting random I/O operations into sequential ones based on data logs, TSUE+ effectively reduces update latency; furthermore, it significantly minimizes recycling overhead using a three-layer log structure and by leveraging the spatio-temporal locality of access patterns. We evaluated TSUE+ and other state of-the-art (SOTA) update mechanisms under diverse encoding schemes, using heterogeneous storage devices—including HDDs, SATA SSDs, NVMe SSDs, and PMEM—and multiple real-world and synthetic workloads: the MSR Cambridge trace, the Alibaba Cloud trace, the Tencent Cloud trace, and multiple synthetic worst-case workloads. Among all the platforms, TSUE+ has achieved significant performance improvements compared to other update methods, it also indicates that TSUE+ can be applied to various storage devices. Additionally, we provided percentile-based tail latency tests and update tests under the worst-case environment, which further demonstrated the ro bustness of TSUE+. Moreover, by enabling prompt log recycle and avoiding unnecessary overwrites and improving update granularity through locality-aware recycling, TSUE+ not only improves update performance but also mitigates write wear on SSD devices, thereby extending their operational lifespan.

Adjoint sensitivity analysis is an exceptionally efficient method for computing the gradient of an objective function with respect to given parameters, playing a crucial role in modern circuit design and verification. According to the principles of the adjoint method, it is necessary to store all essential system state information, such as state vectors and Jacobian matrices, at each time step during the forward integration process in order to construct the adjoint equations during the backward integration. Therefore, the memory overhead of the adjoint method is proportional to the system size and the number of time steps, resulting in prohibitive memory costs for solving large-scale dynamic systems.In this paper, we propose a novel, memory-efficient adjoint sensitivity analysis method that significantly reduces the memory overhead of storing system state information by employing error-bounded lossy compression techniques. Our compression algorithm effectively utilizes the spatiotemporal characteristics of data in circuit simulations and incorporates stringent error control mechanisms. This approach achieves a two-order-ofmagnitude reduction in memory overhead during simulation while ensuring that the accuracy of the adjoint solution remains unaffected.

Mixture-of-Experts (MoE) model parallelism is prevalent in training Large Language Models (e.g., ChatGPT). However, the intensive all-to-all collective communication of the MoE layer’s intermediate computing results substantially degrades MoE training efficiency. In this paper, we propose BirdMoE, a novel load-aware communication compression technique with Bi-random quantization for MoE training with two core modules. Specifically, BirdMoE employs a lightweight Random Quantization (RQ) with expectation invariance property to efficiently map the floating-point intermediate computing results into integers while maintaining the MoE training quality. Additionally, BirdMoE utilizes a Mixed Precision (MP) strategy to dynamically balance the communication loads among expert nodes, significantly improving all-to-all communication efficiency for the MoE training system. Experiments on four typical MoE training tasks demonstrate that BirdMoE achieves higher $4.06 \times- 10.44 \times$ total communication compression ratios and $1.18 \times-5.27 \times$ training speedup compared with the state-of-the-art compression techniques while maintaining the MoE training quality.

Compared to replication-based storage systems, erasure-coded storage incurs significantly higher overhead during data updates. To address this issue, various parity logging methods have been proposed. Nevertheless, due to the long update path and substantial amount of random I/O involved in erasure code update processes, the resulting long latency and low throughput often fail to meet the requirements of high performance applications. To this end, we propose a two-stage data update method called TSUE. TSUE divides the update process into a synchronous stage that records updates in a data log, and an asynchronous stage that recycles the log in real-time. TSUE effectively reduces update latency by transforming random I/O into sequential I/O, and it significantly reduces recycle overhead by utilizing a three-layer log and the spatio-temporal locality of access patterns. In SSDs cluster, TSUE significantly improves update performance, achieving improvements of 7.6× under Ali-Cloud trace, 5× under Ten-Cloud trace, while it also extends the SSD's lifespan by up to 13× through reducing the frequencies of reads/writes and of erase operations.

The row-oriented store model is the cornerstone component of modern online transaction processing (OLTP) database systems. In response to the massive increase in data within database systems, compression techniques are employed to enhance storage efficiency. Regrettably, current compression methods suffer from either the amplification issue due to coarse compression granularity or inefficient decompression operations, thus usually decreasing the speed of query processing. To this end, we present DPTC, a cost-effective and decompression-transparent approach designed to compress data pages, the basic storage unit of OLTP database systems. Specifically, (1) DPTC applies a row-wise decompression-oriented structure to track the first occurrence of redundant data in compressed data, which effectively supports the decompression of individual records from pages, thereby avoiding unwarranted decompression in record access. Moreover, (2) DPTC employs an in-page dynamic packing strategy, which determines the compression units based on the impact of each data reduction operation on the compression gains and eliminates gains-inefficient data reductions. Furthermore, (3) DPTC utilizes a SIMD-based mechanism that leverages the characteristics of operations within the decompression process to improve the decompression speed. Our evaluation results confirm that DPTC is efficient in terms of decompression speed and compression ratio. Within an OLTP database system, DPTC yields throughput improvements of up to 4.28 x in TPC-C and reduces latency by up to 33.3% for data point queries in a row-oriented storage engine.

Key-value (KV) storage is widely used in domains such as big data analytics, AI training, databases, and distributed file systems. Traditional KV systems built on DRAM or disk-based architectures struggle to meet the dual requirements of high throughput and strong data persistence demanded by modern applications. Non-Volatile Memory (NVM), with its combination of high bandwidth and persistence, offers a promising foundation for building high-performance, persistent, and largescale KV stores. Consequently, NVM-based KV storage has gained substantial research and industrial interest in recent years. However, directly porting DRAM-or disk-oriented KV designs to NVM devices often yields suboptimal performance. In Non-Uniform Memory Access (NUMA) architectures, frequent NVM persistence operations and high cross-NUMA access latency significantly limit I/O efficiency. To address these challenges, we propose SaneKV, an NVM-optimized KV store. SaneKV introduces a metadata asynchronous persistence mechanism that reduces I/O latency by aggregating metadata write-back operations, and an adaptive NVM data allocation policy to improve throughput. Experimental results show that for small-sized KV workloads, SaneKV achieves up to 75 % higher write throughput and 40 % higher read throughput than state-of-the-art NVM-based KV stores. For large-sized KV workloads, SaneKV achieves comparable peak throughput while requiring up to 40 % fewer threads to reach saturation and delivers over$2 \times$higher overall performance than prior NVM-based designs, demonstrating superior scalability and resource utilization.

Graph Neural Networks (GNNs) have shown remarkable performance across a variety of graph-related tasks.Recent efforts indicate that GNN performance can be enhanced through more sophisticated strategies, such as employing advanced aggregators, increasing aggregation depth, and utilizing larger sampling rates, etc.While these strategies yield promising results, it also incurs a significantly larger memory footprint that can easily surpass the GPU memory capacity.Micro-batching has emerged as a promising method to mitigate GPU memory bottleneck while preserving model accuracy.Nevertheless, integrating micro-batches into GNN

Multimodal large language models (MLLMs) extend LLMs to handle images, videos, and audio by incorporating feature extractors and projection modules. However, these additional components—combined with complex inference pipelines and heterogeneous workloads—introduce significant inference overhead. Therefore, efficiently serving MLLMs remains a major challenge. Current tightly coupled serving architectures struggle to distinguish between mixed request types or adapt parallelism strategies to different inference stages, leading to increased time-to-first-token (TTFT) and poor resource utilization. To address this, we introduce Elastic Multimodal Parallelism (EMP), a new serving paradigm that elastically adapts to resource heterogeneity across request types and inference stages. Building upon EMP, we develop ElasticMM, an MLLM serving system that (1) separates requests into independent modality groups with dynamic resource allocation via a modality-aware load balancer; (2) decouples inference stages and enables parallelism adjustment and adaptive scaling via elastic partition scheduling; and (3) improves inference efficiency through unified multimodal prefix caching and non-blocking encoding. Experiments on diverse real-world datasets show that ElasticMM outperforms state-of-the-art (SOTA) serving systems, reducing TTFT by up to 4.2$\times$ and achieving 3.2–4.5$\times$ higher throughput while meeting service-level objectives (SLOs).

Scientific images play a crucial role in many experimental sciences; however, the large volumes of data generated present significant challenges. Effective image compression must be fast, achieve high compression ratios, and preserve critical domain-specific features. Existing compressors, such as JPEG and SZ, often distort important textures when operating at high compression ratios. Conversely, AI-based compressors offer superior image quality and higher compression ratios but are significantly slower than traditional methods. To address this trade-off, we developed ViSemZ, a high-performance AI-based compressor specifically designed to preserve visual semantics. Our approach enhances AI compression by integrating sparse encoding with variable-length integer truncation, optimized lossless encoding using bitshuffle and a decoupled lookback prefix-sum, and pipelining techniques to enable efficient data streaming and asynchronous processing. Evaluations on scientific datasets demonstrate that, at comparable compression ratios, ViSemZ achieves performance almost on par with existing AI-based compressors while delivering a 9.6× overall compression speedup. These results effectively bridge the performance gap between traditional and AI-based compression methods.

Deep neural networks (DNNs) increasingly rely on parallel structures to enhance performance and efficiency. However, existing machine learning compilers (MLCs) face challenges in optimizing these structures due to limited parallel fusion scopes and insufficient consideration of intra-operator information. This paper introduces Magneto, a novel framework designed to accelerate parallel structures in DNNs through the co-optimization of parallel operators. By expanding the scope of parallel operator fusion and introducing a dedicated co-tuning algorithm, Magneto unlocks new opportunities for co-optimization. Experimental results demonstrate that Magneto outperforms NVIDIA TensorRT and AMD MIGraphX, achieving speedups of 3.02× and 4.19×, respectively.

Second-order optimization methods have been developed to enhance convergence and generalization in deep neural network (DNN) training compared to first-order methods like Stochastic Gradient Descent (SGD). However, these methods face challenges in distributed settings due to high communication overhead. Gradient compression, a technique commonly used to accelerate communication for first-order approaches, often results in low communication reduction ratios, decreased model accuracy, and/or high compression overhead when applied to second-order methods. To address these limitations, we introduce a novel gradient compression method for second-order optimizers called COMPSO. This method effectively reduces communication costs while preserving the advantages of second-order optimization. COMPSO employs stochastic rounding to maintain accuracy and filters out minor gradients to improve compression ratios. Additionally, we develop GPU optimizations to minimize compression overhead and performance modeling to ensure end-to-end performance gains across various systems. Evaluation of COMPSO on different DNN models shows that it achieves a compression ratio of 22.1×, reduces communication time by 14.2×, and improves overall performance by 1.9×, all without any drop in model accuracy.

Over the last nearly 20 years, lossy compression has become an essential aspect of HPC applications’ data pipelines, allowing them to overcome limitations in storage capacity and bandwidth and, in some cases, increase computational throughput and capacity. However, with the adoption of lossy compression comes the requirement to assess and control the impact lossy compression has on scientific outcomes. In this work, we take a major step forward in describing the state of practice and by characterizing workloads. We examine applications’ needs and compressors’ capabilities across 9 different supercomputing application domains. We present 24 takeaways that provide best practices for applications, operational impacts for facilities achieving compressed data, and gaps in application needs not addressed by production compressors that point towards opportunities for future compression research.

Lossless compression is a classic technique for reducing data storage and transmission requirements. Asymmetric Numeral Systems (ANS) is a high-throughput, high-ratio lossless compression algorithm, but it lacks effective support for multi-byte data and cross-platform compatibility. To address this issue, we propose an Adaptive Data Mapping (ADM) scheme, which maps multi-byte integer data into single-byte space based on the data’s characteristics, improving the compression ratio of ANS while maintaining low encoding redundancy. We also optimize the ADM algorithm and the ANS encoder for GPU and CPU architectures, respectively, and combine them to create an efficient and portable ANS encoding method for multi-byte integer data, called MANS. Experimental results show that MANS improves compression ratios by an average of 1.24 ×, achieves 870.27MB/s throughput on CPUs, and delivers up to 288.45 × and 135.86 × speedups on an NVIDIA A100 and an AMD MI210 GPU compared to the CPU version—demonstrating its efficiency and portability across platforms.

Many scientific applications opt for particles instead of meshes as their basic primitives to model complex systems composed of billions of discrete entities. Such applications span a diverse array of scientific domains, including molecular dynamics, cosmology, computational fluid dynamics, and geology. The scale of the particles in those scientific applications increases substantially thanks to the ever-increasing computational power in high-performance computing (HPC) platforms. However, the actual gains from such increases are often undercut by obstacles in data management systems related to data storage, transfer, and processing. Lossy compression has been widely recognized as a promising solution to enhance scientific data management systems regarding such challenges, although most existing compression solutions are tailored for Cartesian grids and thus have sub-optimal results on discrete particle data. In this paper, we introduce LCP, an innovative lossy compressor designed for particle datasets, offering superior compression quality and higher speed than existing compression solutions. Specifically, our contribution is threefold. (1) We propose LCP-S, an error-bound aware block-wise spatial compressor to efficiently reduce particle data size while satisfying the pre-defined error criteria. This approach is universally applicable to particle data across various domains, eliminating the need for reliance on specific application domain characteristics. (2) We develop LCP, a hybrid compression solution for multi-frame particle data, featuring dynamic method selection and parameter optimization. It aims to maximize compression effectiveness while preserving data quality as much as possible by utilizing both spatial and temporal domains. (3) We evaluate our solution alongside eight state-of-the-art alternatives on eight real-world particle datasets from seven distinct domains. The results demonstrate that our solution achieves up to 104% improvement in compression ratios and up to 593% increase in speed compared to the second-best option, under the same error criteria.

Parallel structures have become a key pattern in deep neural networks (DNNs), offering improved efficiency and scalability. However, existing machine learning compilers (MLCs) face challenges in optimizing these structures due to limited parallel fusion scope and insufficient analysis of intra-operator characteristics. This article introduces Magneto, a framework designed to accelerate DNN inference by co-optimizing parallel operators. Magneto broadens the fusion scope and incorporates a specialized co-tuning algorithm to optimize operators jointly. Our approach addresses the unique challenges inherent in optimizing parallel structures, enabling significant performance improvements across various hardware platforms. Experimental results show that Magneto outperforms state-of-the-art NVIDIA TensorRT and AMD MIGraphX, achieving geometric mean speedups of 2.27× and 2.88×, respectively.

Cross-Silo federated learning is widely used for scaling deep neural network (DNN) training over data silos from different locations worldwide while guaranteeing data privacy. Communication has been identified as the main bottleneck when training large-scale models due to large-volume model parameters and gradient transmission across public networks with limited bandwidth. Most previous works focus on gradient compression, while limited work tries to compress parameters that can not be ignored and extremely affect communication performance during the training. To bridge this gap, we proposeFedCSpc: an efficient cross-silo federated learning system with an XAI-driven adaptive parameter compression strategy for large-scale model training. Our work substantially differs from existing gradient compression techniques due to the distinct data features of gradient and parameter. The key contributions of this paper are fourfold. (1) Our designedFedCSpcproposes to compress the parameter during the training using the state-of-the-art error-bounded lossy compressor – SZ3. (2) We develop an adaptive compression error bound adjustment algorithm to guarantee the model accuracy effectively. (3) We exploit an efficient approach to utilize the idle CPU resources of clients to compress the parameters. (4) We perform a comprehensive evaluation with a wide range of models and benchmarks on a GPU cluster with 65 GPUs. Results show thatFedCSpccan achieve the same model accuracy as FedAvg while reducing the data volume of parameters and gradients in communication by up to 7.39× and 288×, respectively. With 32 clients on a 4Gb size model,FedCSpcsignificantly outperforms FedAvg in wall-clock time in the emulated WAN environment (at the bandwidth of 1 Gbps or lower without loss of generality).

Emerging deep learning workloads urgently need fast general matrix multiplication (GEMM). Thus, one of the critical features of machine-learning-specific accelerators such as NVIDIA Tensor Cores, AMD Matrix Cores, and Google TPUs is the support of mixed-precision enabled GEMM. For DNN models, lower-precision FP data formats and computation offer acceptable correctness but significant performance, area, and memory footprint improvement. While promising, the mixed-precision computation on error resilience remains unexplored. To this end, we develop a fault injection framework that systematically injects fault into the mixed-precision computation results. We investigate how the faults affect the accuracy of machine learning applications. Based on error resilience characteristics, we offer lightweight error detection and correction solutions that significantly improve the overall model accuracy by 75% if the models experience hardware faults. The solutions can be efficiently integrated into the accelerator's pipelines.

Adjoint sensitivity analysis is critical in modern integrated circuit design and verification, but its computational intensity grows significantly with the circuit size, the number of objective functions, and the accumulation of time points. This growth can impede its wider application. The intimate link between the forward integration in transient analysis and the reverse integration in adjoint sensitivity analysis allows for the retention of Jacobian matrices from transient analysis, thereby speeding up sensitivity analysis. However, Jacobian matrices across multiple timesteps are often so large that they cannot be stored in memory during the forward integration process, necessitating disk storage and incurring significant I/O overhead. To address this, we develop a memory-efficient sensitivity analysis method that utilizes data compression to minimize memory overhead during simulation and enhance analysis efficiency. Our compression method can efficiently compress the sparse tensor that contains the Jacobian matrices over time by exploiting the spatiotemporal characteristics of the data and circuit attributes. It also introduces a shared-indices technique, a cutting-edge spatiotemporal prediction model, and robust residual encoding. We evaluate our compression method on 7 datasets from real-world simulations and demonstrate that it can reduce memory requirements by more than 16x on average, which is significantly more efficient than other state-of-the-art compression techniques.

As time series data become popular, their volume increases rapidly. Time series databases are designed for such data, and they process data in short slices, meaning that the compression units for compressors are small. How to compress the short slices of floating-points while reserving a high compression ratio and a high decompression speed remains a problem.To solve the problem, we propose a lossy compressor Machete. It uses an efficient hybrid encoder of Huffman encoding and variable length quantity (VLQ). Adaptive encoding selection makes it excel on short-slice data compression ratio, while the simple framework ensures fast decompression. We also find a limitation in VLQ and propose the optimal VLQ to further improve the compression ratio.Our evaluation on four real-world datasets shows that Machete outperforms state-of-the-art compressors by 32%−80% on compression ratio and achieves the fastest decompression speed on two datasets. When applied to a well-known time series database InfluxDB, Machete saves disk usage up to 79% and improves the query performance of the InfluxDB database by saving I/O.

Lossy compression and asynchronous I/O are two of the most effective solutions for reducing storage overhead and enhancing I/O performance in large-scale high-performance computing (HPC) applications. However, current approaches have limitations that prevent them from fully leveraging lossy compression, and they may also result in task collisions, which restrict the overall performance of HPC applications. To address these issues, we propose an optimization approach for the task scheduling problem that encompasses computation, compression, and I/O. Our algorithm adaptively selects the optimal compression and I/O queue to minimize the performance degradation of the computation. We also introduce an intra-node I/O workload balancing mechanism that evenly distributes the workload across different processes. Additionally, we design a framework that incorporates fine-grained compression, a compressed data buffer, and a shared Huffman tree to fully benefit from our proposed task scheduling. Experimental results with up to 16 nodes and 64 GPUs from ORNL Summit, as well as real-world HPC applications, demonstrate that our solution reduces I/O overhead by up to 3.8× and 2.6× compared to non-compression and asynchronous I/O solutions, respectively.

Recent years have witnessed a clear trend towards language models with an ever-increasing number of parameters, as well as the growing training overhead and memory usage. Distributed training, particularly through Sharded Data Parallelism (ShardedDP) which partitions optimizer states among workers, has emerged as a crucial technique to mitigate training time and memory usage. Yet, a major challenge in the scalability of ShardedDP is the intensive communication of weights and gradients. While compression techniques can alleviate this issue, they often result in worse accuracy. Driven by this limitation, we propose SDP4Bit (Toward 4Bit Communication Quantization in Sharded Data Parallelism for LLM Training), which effectively reduces the communication of weights and gradients to nearly 4 bits via two novel techniques: quantization on weight differences, and two-level gradient smooth quantization. Furthermore, SDP4Bit presents an algorithm-system co-design with runtime optimization to minimize the computation overhead of compression. Additional to the theoretical guarantees of convergence, we empirically evaluate the accuracy of SDP4Bit on the pre-training of GPT models with up to 6.7 billion parameters, and the results demonstrate a negligible impact on training loss. Furthermore, speed experiments show that SDP4Bit achieves up to 4.08× speedup in end-to-end throughput on a scale of 128 GPUs.

Error-bounded lossy compression is a critical technique for significantly reducing scientific data volumes. Compared to CPU-based compressors, GPU-based compressors exhibit substantially higher throughputs, fitting better for today’s HPC applications. However, the critical limitations of existing GPU-based compressors are their low compression ratios and qualities, severely restricting their applicability. To overcome these, we introduce a new GPU-based error-bounded scientific lossy compressor named CUSZ-i, with the following contributions: (1) A novel GPU-optimized interpolation-based prediction method significantly improves the compression ratio and decompression data quality. (2) The Huffman encoding module in CUSZ-i is optimized for better efficiency. (3) CUSZ-i is the first to integrate the NVIDIA Bitcomp-lossless as an additional compression-ratio-enhancing module. Evaluations show that CUSZ-i significantly outperforms other latest GPU-based lossy compressors in compression ratio under the same error bound (hence, the desired quality), showcasing a 476% advantage over the second-best. This leads to CUSZ-i’s optimized performance in several real-world use cases.

DLRM is a state-of-the-art recommendation system model that has gained widespread adoption across various industry applications. The large size of DLRM models, however, necessitates the use of multiple devices/GPUs for efficient training. A significant bottleneck in this process is the time-consuming all-to-all communication required to collect embedding data from all devices. To mitigate this, we introduce a method that employs error-bounded lossy compression to reduce the communication data size and accelerate DLRM training. We develop a novel error-bounded lossy compression algorithm, informed by an in-depth analysis of embedding data features, to achieve high compression ratios. Moreover, we introduce a dual-level adaptive strategy for error-bound adjustment, spanning both table-wise and iteration-wise aspects, to balance the compression benefits with the potential impacts on accuracy. We further optimize our compressor for PyTorch tensors on GPUs, minimizing compression overhead. Evaluation shows that our method achieves a 1.38 × training speedup with a minimal accuracy impact.

Multi-resolution methods such as Adaptive Mesh Refinement (AMR) can enhance storage efficiency for HPC applications generating vast volumes of data. However, their applicability is limited and cannot be universally deployed across all applications. Furthermore, integrating lossy compression with multi-resolution techniques to further boost storage efficiency encounters significant barriers. To this end, we introduce an innovative workflow that facilitates high-quality multi-resolution data compression for both uniform and AMR simulations. Initially, to extend the usability of multi-resolution techniques, our workflow employs a compression-oriented Region of Interest (ROI) extraction method, transforming uniform data into a multi-resolution format. Subsequently, to bridge the gap between multi-resolution techniques and lossy compressors, we optimize three distinct compressors, ensuring their optimal performance on multi-resolution data. These optimizations can improve the compression ratio of SOTA approaches by up to $3.3 \times$ under the same data quality loss. Lastly, we incorporate an advanced uncertainty visualization method into our workflow to understand the potential impacts of lossy compression. Experimental evaluation demonstrates that our workflow achieves significant compression quality improvements.

While both the database and high-performance computing (HPC) communities utilize lossless compression methods to minimize floating-point data size, a disconnect persists between them. Each community designs and assesses methods in a domain-specific manner, making it unclear if HPC compression techniques can benefit database applications or vice versa. With the HPC community increasingly leaning towards in-situ analysis and visualization, more floating-point data from scientific simulations are being stored in databases like Key-Value Stores and queried using in-memory retrieval paradigms. This trend underscores the urgent need for a collective study of these compression methods' strengths and limitations, not only based on their performance in compressing data from various domains but also on their runtime characteristics. Our study extensively evaluates the performance of eight CPU-based and five GPU-based compression methods developed by both communities, using 33 real-world datasets assembled in the Floating-point Compressor Benchmark (FCBench). Additionally, we utilize the roofline model to profile their runtime bottlenecks. Our goal is to offer insights into these compression methods that could assist researchers in selecting existing methods or developing new ones for integrated database and HPC applications.

Today's scientific simulations require significant data volume reduction because of the enormous amounts of data produced and the limited I/O bandwidth and storage space. Error-bounded lossy compression has been considered one of the most effective solutions to the above problem. However, little work has been done to improve error-bounded lossy compression for Adaptive Mesh Refinement (AMR) simulation data. Unlike the previous work that only leverages 1D compression, in this work, we propose an approach (TAC) to leverage high-dimensional SZ compression for each refinement level of AMR data. To remove the data redundancy across different levels, we propose several pre-process strategies and adaptively use them based on the data features. We further optimizeTACtoTAC+by improving the lossless encoding stage of SZ compression to handle many small AMR data blocks after the pre-processing efficiently. Experiments on 10 AMR datasets from three real-world large-scale AMR simulations demonstrate thatTAC+can improve the compression ratio by up to 4.9× under the same data distortion, compared to the state-of-the-art method. In addition, we leverage the flexibility of our approach to tune the error bound for each level, which achieves much lower data distortion on two application-specific metrics.

The Top-K sparsification-based compression framework is extensively explored for reducing communication costs in distributed learning. However, we identified several issues with existing Top-K sparsification-based compression methods: (i) The limited compressibility of the Top-K parameter's indexes critically restricts the overall communication compression ratio; (ii) Several time-consuming compression operations significantly offset the benefits of communication compression; (iii) The use of error feedback techniques to maintain model quality results in a high memory footprint consumption. To solve these issues, we propose BIRD, a lightweight tensor-wiseBi-Random samplingstrategy with an expectation invariance property. Specifically, BIRD applies a tensor-wiseindex sharingmechanism that reduces the index proportion by allowing multiple tensor elements to share a single index, thus improving the overall compression ratio. Additionally, BIRD replaces the time-consuming Top-K sorting with a fasterBi-Random samplingstrategy based on the aforementionedindex sharingmechanism, significantly reducing compression overheads; Moreover, BIRD establishes anexpectation invarianceproperty into theBi-Random samplingto ensure an approximate unbiased representation for the$L_1$-norm of the sampled tensors, effectively maintaining the model quality without incurring extra memory costs. We further optimize BIRD to BIRD+ by introducing the uniform distribution-based sampling and Gamma correction on the tensor-wise sampling process, achieving a more flexibly adjustment of the sparsity with better convergence performance. Experimental evaluations across multiple conventional distributed learning tasks demonstrate that compared to state-of-the-art approaches, BIRD+ achieves higher communication compression ratios up to 36.2$\times$and higher computation throughput up to 149.6$\times$while maintaining the model quality without incurring extra memory costs.

Low-rank compression is an important model compression strategy for obtaining compact neural network models. In general, because the rank values directly determine the model complexity and model accuracy, proper selection of layer-wise rank is very critical and desired. To date, though many low-rank compression approaches, either selecting the ranks in a manual or automatic way, have been proposed, they suffer from costly manual trials or unsatisfied compression performance. In addition, all of the existing works are not designed in a hardware-aware way, limiting the practical performance of the compressed models on real-world hardware platforms. To address these challenges, in this paper we propose HALOC, a hardware-aware automatic low-rank compression framework. By interpreting automatic rank selection from an architecture search perspective, we develop an end-to-end solution to determine the suitable layer-wise ranks in a differentiable and hardware-aware way. We further propose design principles and mitigation strategy to efficiently explore the rank space and reduce the potential interference problem. Experimental results on different datasets and hardware platforms demonstrate the effectiveness of our proposed approach. On CIFAR-10 dataset, HALOC enables 0.07% and 0.38% accuracy increase over the uncompressed ResNet-20 and VGG-16 models with 72.20% and 86.44% fewer FLOPs, respectively. On ImageNet dataset, HALOC achieves 0.9% higher top-1 accuracy than the original ResNet-18 model with 66.16% fewer FLOPs. HALOC also shows 0.66% higher top-1 accuracy increase than the state-of-the-art automatic low-rank compression solution with fewer computational and memory costs. In addition, HALOC demonstrates the practical speedups on different hardware platforms, verified by the measurement results on desktop GPU, embedded GPU and ASIC accelerator.

Today's large-scale scientific applications running on high-performance computing (HPC) systems generate vast data volumes. Thus, data compression is becoming a critical technique to mitigate the storage burden and data-movement cost. However, existing lossy compressors for scientific data cannot achieve a high compression ratio and throughput simultaneously, hindering their adoption in many applications requiring fast compression, such as in-memory compression. To this end, in this work, we develop a fast and high- ratio error-bounded lossy compressor on GPUs for scientific data (called FZ-GPU). Specifically, we first design a new compression pipeline that consists of fully parallelized quantization, bitshuffle, and our newly designed fast encoding. Then, we propose a series of deep architectural optimizations for each kernel in the pipeline to take full advantage of CUDA architectures. We propose a warp-level optimization to avoid data conflicts for bit-wise operations in bitshuffle, maximize shared memory utilization, and eliminate unnecessary data movements by fusing different compression kernels. Finally, we evaluate FZ-GPU on two NVIDIA GPUs (i.e., A100 and RTX A4000) using six representative scientific datasets from SDRBench. Results on the A100 GPU show that FZ-GPU achieves an average speedup of 4.2× over cuSZ and an average speedup of 37.0× over a multi-threaded CPU implementation of our algorithm under the same error bound. FZ-GPU also achieves an average speedup of 2.3× and an average compression ratio improvement of 2.0× over cuZFP under the same data distortion.

Collaborative filtering (CF) has been proven to be one of the most effective techniques for recommendation. Among all CF approaches, SimpleX is the state-of-the-art method that adopts a novel loss function and a proper number of negative samples. However, there is no work that optimizes SimpleX on multi-core CPUs, leading to limited performance. To this end, we perform an in-depth profiling and analysis of existing SimpleX implementations and identify their performance bottlenecks including (1) irregular memory accesses, (2) unnecessary memory copies, and (3) redundant computations. To address these issues, we propose an efficient CF training system (called HEAT) that fully enables the multi-level caching and multi-threading capabilities of modern CPUs. Specifically, the optimization of HEAT is threefold: (1) It tiles the embedding matrix to increase data locality and reduce cache misses (thus reduces read latency); (2) It optimizes stochastic gradient descent (SGD) with sampling by parallelizing vector products instead of matrix-matrix multiplications, in particular the similarity computation therein, to avoid memory copies for matrix data preparation; and (3) It aggressively reuses intermediate results from the forward phase in the backward phase to alleviate redundant computation. Evaluation on five widely used datasets with both x86- and ARM-architecture processors shows that HEAT achieves up to 45.2× speedup over existing CPU solution and 4.5× speedup and 7.9× cost reduction in Cloud over existing GPU solution with NVIDIA V100 GPU.

Deep Learning Recommendation Models (DLRMs) are important applications in various domains and have evolved into one of the largest and most important machine learning applications. With their trillions of parameters necessarily exceeding the high bandwidth memory (HBM) capacity of GPUs, ever more massive DLRMs require large-scale multi-node systems for distributed training and inference. However, these all suffer from the all-to-all communication bottleneck, which limits scalability.

Today's graphics processing unit (GPU) applications produce vast volumes of data, which are challenging to store and transfer efficiently. Thus, data compression is becoming a critical technique to mitigate the storage burden and communication cost. LZSS is the core algorithm in many widely used compressors, such as Deflate. However, existing GPU-based LZSS compressors suffer from low throughput due to the sequential nature of the LZSS algorithm. Moreover, many GPU applications produce multi-byte data (e.g., int16/int32 index, floating-point numbers), while the current LZSS compression only takes single-byte data as input. To this end, in this work, we propose gpuLZ, a highly efficient LZSS compression on modern GPUs for multi-byte data. The contribution of our work is fourfold: First, we perform an in-depth analysis of existing LZ compressors for GPUs and investigate their main issues. Then, we propose two main algorithm-level optimizations. Specifically, we (1) change prefix sum from one pass to two passes and fuse multiple kernels to reduce data movement between shared memory and global memory, and (2) optimize existing pattern-matching approach for multi-byte symbols to reduce computation complexity and explore longer repeated patterns. Third, we perform architectural performance optimizations, such as maximizing shared memory utilization by adapting data partitions to different GPU architectures. Finally, we evaluate gpuLZ on six datasets of various types with NVIDIA A100 and A4000 GPUs. Results show that gpuLZ achieves up to 272.1× speedup on A4000 and up to 1.4× higher compression ratio compared to state-of-the-art solutions.

Tucker decomposition is one of the SOTA CNN model compression techniques. However, unlike the FLOPs reduction, we observe very limited inference time reduction with Tucker-compressed models using existing GPU software such as cuDNN. To this end, we propose an efficient end-to-end framework that can generate highly accurate and compact CNN models via Tucker decomposition and optimized inference code on GPUs. Specifically, we propose an ADMM-based training algorithm that can achieve highly accurate Tucker-format models. We also develop a high-performance kernel for Tucker-format convolutions and analytical performance models to guide the selection of execution parameters. We further propose a co-design framework to determine the proper Tucker ranks driven by practical inference time (rather than FLOPs). Our evaluation on five modern CNNs with A100 demonstrates that our compressed models with our optimized code achieve up to 2.21× speedup over cuDNN, 1.12× speedup over TVM, and 3.27× over the original models using cuDNN with at most 0.05% accuracy loss.

Soft errors are prevalent in modern High-Performance Computing (HPC) systems, resulting in silent data corruptions (SDCs), compromising system reliability. Instruction duplication is a widely used software-based protection technique against SDCs. Existing instruction duplication techniques are mostly implemented at LLVM level and may suffer from low SDC coverage at assembly level. In this paper, we evaluate instruction duplication at both LLVM and assembly levels. Our study shows that existing instruction duplication techniques have protection deficiency at assembly level and are usually over-optimistic in the protection. We investigate the root-causes of the protection deficiency and propose a mitigation technique, Flowery, to solve the problem. Our evaluation shows that Flowery can effectively protect programs from SDCs evaluated at assembly level.

As supercomputers advance towards exascale capabilities, computational intensity increases significantly, and the volume of data requiring storage and transmission experiences exponential growth. Adaptive Mesh Refinement (AMR) has emerged as an effective solution to address these two challenges. Concurrently, error-bounded lossy compression is recognized as one of the most efficient approaches to tackle the latter issue. Despite their respective advantages, few attempts have been made to investigate how AMR and error-bounded lossy compression can function together. To this end, this study presents a novel in-situ lossy compression framework that employs the HDF5 filter to improve both I/O costs and boost compression quality for AMR applications. We implement our solution into the AMReX framework and evaluate on two real-world AMR applications, Nyx and WarpX, on the Summit supercomputer. Experiments with 4096 CPU cores demonstrate that AMRIC improves the compression ratio by up to 81× and the I/O performance by up to 39× over AMReX's original compression solution.

Today's scientific simulations require a significant reduction of data volume because of extremely large amounts of data they produce and the limited I/O bandwidth and storage space. Error-bounded lossy compression has been considered one of the most effective solutions to the above problem. In practice, however, the best-fit compression method often needs to be customized or optimized in particular because of diverse characteristics in different datasets and various user requirements on the compression quality and performance. In this paper, we address this issue with a novel modular, composable compression framework named SZ3. Our contributions are four-folds. (1) We develop SZ3 which features an innovative modular abstraction for the prediction-based compression framework, such that compression modules can be plugged in easily to create new compressors based on characteristics of data and user requirements. (2) We create a new compression pipeline by SZ3 for GAMESS data, which significantly improves the compression ratios over state-of-the-art compressors. (3) We develop an adaptive compression pipeline by SZ3 for APS data with minimal efforts, which leads to the best rate-distortion among all existing error-bounded lossy compressors for any bit-rate. (4) We compare the sustainability of SZ3 with leading error-bounded prediction-based compressors, and then demonstrate the necessity of diverse pipelines by integrating and evaluating several compression pipelines on diverse scientific datasets from multiple disciplines. Experiments show that SZ3 incurs very limited overhead in compressor integration and our customized compression pipelines lead to up to 20% improvement in compression ratios under the same data distortion, when compared with the best existing approach.

Deep neural networks (DNNs) have achieved remarkable success in many fields. However, large-scale DNNs also bring storage costs when storing snapshots for preventing clusters’ frequent failures or incur significant communication overheads when transmitting DNNs in the Federated Learning (FL). Recently, several approaches, such as Delta-DNN and LC-Checkpoint, aim to reduce the size of DNNs’ snapshot storage by compressing the difference between two neighboring versions of the DNNs (a.k.a., delta). However, we observe that existing approaches, applying traditional global lossy quantization techniques in DNN's delta compression, can not fully exploit the data similarity since the parameters’ value ranges vary among layers. To fully explore the similarity of the delta model and improve the compression ratio, we propose a quantization-based local-sensitive delta compression approach, named QD-Compressor, by developing a layer-based local-sensitive quantization scheme and error feedback mechanism. Specifically, the quantizers and number of quantization bits are adaptive among layers based on the value distribution and weighted entropy of the delta's parameters. To avoid quantization error degrading the performance of the restored model, an alternative error feedback mechanism is designed to dynamically correct the quantization error during the training process. Experiments on multiple popular DNNs and datasets show that QD-Compressor obtains a higher 7×-40× compression ratio in the model snapshot compression scenario than the state-of-the-art approaches. Additionally, QD-Compressor achieves an 11×-15× compression ratio to the residual model of the Federated Learning compression scenario.

Influence maximization aims to select k most-influential vertices or seeds in a network, where influence is defined by a given diffusion process. Although computing optimal seed set is NP-Hard, efficient approximation algorithms exist. However, even state-of-the-art parallel implementations are limited by a sampling step that incurs large memory footprints. This in turn limits the problem size reach and approximation quality. In this work, we study the memory footprint of the sampling process collecting reverse reachability information in the IMM (Influence Maximization via Martingales) algorithm over large real-world social networks. We present a memory-efficient optimization approach (called HBMax) based on Ripples, a state-of-the-art multi-threaded parallel influence maximization solution. Our approach, HBMax, uses a portion of the reverse reachable (RR) sets collected by the algorithm to learn the characteristics of the graph. Then, it compresses the intermediate reverse reachability information with Huffman coding or bitmap coding, and queries on the partially decoded data, or directly on the compressed data to preserve the memory savings obtained through compression. Considering a NUMA architecture, we scale up our solution on 64 CPU cores and reduce the memory footprint by up to 82.1% with average 6.3% speedup (encoding overhead is offset by performance gain from memory reduction) without loss of accuracy. For the largest tested graph Twitter7 (with 1.4 billion edges), HBMax achieves 5.9× compression ratio and 2.2× speedup.

Graph Neural Networks (GNNs) have drawn tremendous attention due to their unique capability to extend Machine Learning (ML) approaches to applications broadly-defined as having unstructured data, especially graphs. Compared with other Machine Learning (ML) modalities, the acceleration of Graph Neural Networks (GNNs) is more challenging due to the irregularity and heterogeneity derived from graph typologies. Existing efforts, however, have focused mainly on handling graphs' irregularity and have not studied their heterogeneity. To this end we propose H-GCN, a PL (Programmable Logic) and AIE (AI Engine) based hybrid accelerator that leverages the emerging heterogeneity of Xilinx Versal Adaptive Compute Acceleration Platforms (ACAPs) to achieve high-performance GNN inference. In particular, H-GCN partitions each graph into three subgraphs based on its inherent heterogeneity, and processes them using PL and AIE, respectively. To further improve performance, we explore the sparsity support of AIE and develop an efficient density-aware method to automatically map tiles of sparse matrix-matrix multiplication (SpMM) onto the systolic tensor array. Compared with state-of-the-art GCN accelerators, H-GCN achieves, on average, speedups of 1.1~2.3x.

Today's scientific simulations require a significant reduction of data volume because of extremely large amounts of data they produce and the limited I/O bandwidth and storage space. Error-bounded lossy compression has been considered one of the most effective solutions to the above problem. However, little work has been done to improve error-bounded lossy compression for Adaptive Mesh Refinement (AMR) simulation data. Unlike the previous work that only leverages 1D compression, in this work, we propose to leverage high-dimensional (e.g., 3D) compression for each refinement level of AMR data. To remove the data redundancy across different levels, we propose three pre-process strategies and adaptively use them based on the data characteristics. Experiments on seven AMR datasets from a real-world large-scale AMR simulation demonstrate that our proposed approach can improve the compression ratio by up to 3.3X under the same data distortion, compared to the state-of-the-art method. In addition, we leverage the flexibility of our approach to tune the error bound for each level, which achieves much lower data distortion on two application-specific metrics.

Today's scientific high-performance computing applications and advanced instruments are producing vast volumes of data across a wide range of domains, which impose a serious burden on data transfer and storage. Error-bounded lossy compression has been developed and widely used in the scientific community because it not only can significantly reduce the data volumes but also can strictly control the data distortion based on the user-specified error bound. Existing lossy compressors, however, cannot offer ultrafast compression speed, which is highly demanded by numerous applications or use cases (such as in-memory compression and online instrument data compression). In this paper, we propose a novel ultrafast error-bounded lossy compressor that can obtain fairly high compression performance on both CPUs and GPUs and with reasonably high compression ratios. The key contributions are threefold. (1) We propose a generic error-bounded lossy compression framework---called SZx---that achieves ultrafast performance through its novel design comprising only lightweight operations such as bitwise and addition/subtraction operations, while still keeping a high compression ratio. (2) We implement SZx on both CPUs and GPUs and optimize the performance according to their architectures. (3) We perform a comprehensive evaluation with six real-world production-level scientific datasets on both CPUs and GPUs. Experiments show that SZx is 2~16x faster than the second-fastest existing error-bounded lossy compressor (either SZ or ZFP) on CPUs and GPUs, with respect to both compression and decompression.

Error-bounded lossy compression is one of the most effective techniques for reducing scientific data sizes. However, the traditional trial-and-error approach used to configure lossy compressors for finding the optimal trade-off between reconstructed data quality and compression ratio is prohibitively expensive. To resolve this issue, we develop a general-purpose analytical ratio-quality model based on the prediction-based lossy compression framework, which can effectively foresee the reduced data quality and compression ratio, as well as the impact of lossy compressed data on post-hoc analysis quality. Our analytical model significantly improves the prediction-based lossy compression in three use-cases: (1) optimization of predictor by selecting the best-fit predictor; (2) memory compression with a target ratio; and (3) in-situ compression optimization by fine-grained tuning error-bounds for various data partitions. We evaluate our analytical model on 10 scientific datasets, demonstrating its high accuracy (93.47% accuracy on average) and low computational cost (up to 18.7x lower than the trial-and-error approach) for estimating the compression ratio and the impact of lossy compression on post-hoc analysis quality. We also verify the high efficiency of our ratio-quality model using different applications across the three use-cases. In addition, our experiment demonstrates that our modeling-based approach reduces the time to store the 3D RTM data with HDF5 by up to 3.4 x with 128 CPU cores over the traditional solution.

Uncertain or probabilistic graphs have been ubiquitously used to represent noisy, incomplete, and inaccurate linked data in many emerging big-data mining and analytics applications. It is impractical to solve uncertain graph problems exactly as it requires to evaluate an exponential number of certain instances (or "possible worlds") generated from an uncertain graph. Previously, several CPU-based techniques were proposed to use sampling for uncertain graph processing. However, we observe that (1) they suffer from low computation efficiency and large memory overhead due to unnecessary edge sampling at runtime; (2) they cannot leverage the massive parallelism provided by modern general-purpose accelerators; and (3) there lacks a general programming framework for high-performance uncertain graph processing. To tackle these challenges, we propose a novel runtime path sampling method, which is able to identify and eliminate unnecessary edge sampling via incremental path identification and filtering, resulting in significant reduction in computation and data movement. Centered around this idea, we introduce a general uncertain graph processing framework for multi-GPU systems, named BPGraph1. BPGraph provides general support for users to design and optimize a wide-range of uncertain graph algorithms and applications without concerning about the underlying complexity. Extensive evaluation on a variety of real-world uncertain graph applications demonstrates an average speedup of 26X (up to 43X) and better scalability from BPGraph over the state-of-the-art frameworks.

As HPC systems continue to grow to exascale, the amount of data that needs to be saved or transmitted is exploding. To this end, many previous works have studied using error-bounded lossy compressors to reduce the data size and improve the I/O performance. However, little work has been done for effectively offloading lossy compression onto FPGA-based SmartNICs to reduce the compression overhead. In this paper, we propose a hardware-algorithm co-design for an efficient and adaptive lossy compressor for scientific data on FPGAs (called CEAZ), which is the first lossy compressor that can achieve high compression ratios and throughputs simultaneously. Specifically, we propose an efficient Huffman coding approach that can adaptively update Huffman codewords online based on codewords generated offline, from a variety of representative scientific datasets. Moreover, we derive a theoretical analysis to support a precise control of compression ratio under an error-bounded compression mode, enabling accurate offline Huffman codewords generation. This also helps us create a fixed-ratio compression mode for consistent throughput. In addition, we develop an efficient compression pipeline by adopting cuSZ's dual-quantization algorithm to our hardware use cases. Finally, we evaluate CEAZ on five real-world datasets with both a single FPGA board and 128 nodes (to accelerate parallel I/O). Experiments show that CEAZ outperforms the second-best FPGA-based lossy compressor by 2.3X of throughput and 3.0X of ratio. It also improves MPI_File_write and MPI_Gather throughputs by up to 28.9X and 37.8X, respectively.

More and more HPC applications require fast and effective compression techniques to handle large volumes of data in storage and transmission. Not only do these applications need to compress the data effectively during simulation, but they also need to perform decompression efficiently for post hoc analysis. SZ is an error-bounded lossy compressor for scientific data, and cuSZ is a version of SZ designed to take advantage of the GPU's power. At present, cuSZ's compression performance has been optimized significantly while its decompression still suffers considerably lower performance because of its sophisticated loss-less compression step-a customized Huffman decoding. In this work, we aim to significantly improve the Huffman decoding performance for cuSZ, thus improving the overall decompression performance in turn. To this end, we first investigate two state-of-the-art GPU Huffman decoders in depth. Then, we propose a deep architectural optimization for both algorithms. Specifically, we take full advantage of CUDA GPU architectures by using shared memory on decoding/writing phases, online tuning the amount of shared memory to use, improving memory access patterns, and reducing warp divergence. Finally, we evaluate our optimized decoders on an Nvidia V100 GPU using eight representative scientific datasets. Our new decoding solution obtains an average speedup of 3.64× over cuSZ's Huffman decoder and improves its overall decompression performance by 2.43× on average.

Modern HPC applications produce increasingly large amounts of data, which limits the performance of current extreme-scale systems. Lossy compression, helps to mitigate this issue by decreasing the size of data generated by these applications. SZ, a current state-of-the-art lossy compressor, is able to achieve high compression ratios, but its prediction/quantization methods contain RAW dependencies that prevent parallelizing this step of the compression. Recent work proposes a parallel dual prediction/quantization algorithm for GPUs which removes these dependencies. However, some HPC systems and applications do not use GPUs, and could still benefit from the fine-grained parallelism of this method. Using the dual-quantization technique, we implement and optimize a SIMD vectorized CPU version of SZ (vecSZ), and create a heuristic for selecting the optimal block size and vector length. We propose a novel block padding algorithm to decrease the number of unpredictable values along compression block borders and find it reduces the number of prediction outliers by up to 100%. We measure performance of our vecSZ against an CPU version of SZ using dual-quantization, pSZ, as well as SZ-1.4. Using real-world scientific datasets, we evaluate vecSZ on the Intel Skylake and AMD Rome architectures. vecSZ results in up to 32% improvement in rate-distortion and up to 15× speedup over SZ-1.4, achieving a prediction and quantization bandwidth in excess of 3.4 GB/s.

Lossy compression is one of the most efficient solutions to reduce storage overhead and improve I/O performance for HPC applications. However, existing parallel I/O libraries cannot fully utilize lossy compression to accelerate parallel write due to the lack of deep understanding on compression-write performance. To this end, we propose to deeply integrate predictive lossy compression with HDF5 to significantly improve the parallel-write performance. Specifically, we propose analytical models to predict the time of compression and parallel write before the actual compression to enable compression-write overlapping. We also introduce an extra space in the process to handle possible data overflows resulting from prediction uncertainty in compression ratios. Moreover, we propose an optimization to reorder the compression tasks to increase the overlapping efficiency. Experiments with up to 4,096 cores from Summit show that our solution improves the write performance by up to$4.5\times$and$2.9\times$over the non-compression and lossy compression solutions, respectively, with only 1.5% storage overhead (compared to original data) on two real-world HPC applications.

Today's scientific simulations and instruments are producing a large amount of data, leading to difficulties in storing, transmitting, and analyzing these data. While error-controlled lossy compressors are effective in significantly reducing data volumes and efficiently developing databases for multiple scientific applications, they mainly support error controls on raw data, which leaves a significant gap between the data and user's downstream analysis. This may cause unqualified uncertainties in the outcomes of the analysis, a.k.a quantities of interest (QoIs), which are the major concerns of users in adopting lossy compression in practice. In this paper, we propose rigorous mathematical theories to preserve four families of QoIs that are widely used in scientific analysis during lossy compression along with practical implementations. Specifically, we first develop the error control theory for univariate QoIs which are essential for computing physical properties such as kinetic energy, followed by multivariate QoIs that are more commonly used in real-world applications. The proposed method is integrated into a state-of-the-art compression framework in a modular fashion, which could easily adapt to new QoIs and new compression algorithms. Experiments on real-world datasets demonstrate that the proposed method provides faithful error control on important QoIs including kinetic energy, regional average, and isosurface without trials and errors, while offering compression ratios that are up to 4X of the compression ratios provided by state-of-the-art compressors.

Nowadays, data reduction is becoming increasingly important in dealing with the large amounts of scientific data. Existing multilevel compression algorithms offer a promising way to manage scientific data at scale, but may suffer from relatively low performance and reduction quality. In this paper, we propose MGARD+, a multilevel data reduction and refactoring framework drawing on previous multilevel methods, to achieve high-performance data decomposition and high-quality error-bounded lossy compression. Our contributions are four-fold: 1) We propose to leverage a level-wise coefficient quantization method, which uses different error tolerances to quantize the multilevel coefficients. 2) We propose an adaptive decomposition method which treats the multilevel decomposition as a preconditioner and terminates the decomposition process at an appropriate level. 3) We leverage a set of algorithmic optimization strategies to significantly improve the performance of multilevel decomposition/recomposition. 4) We evaluate our proposed method using four real-world scientific datasets and compare with several state-of-the-art lossy compressors. Experiments demonstrate that our optimizations improve the decomposition/recomposition performance of the existing multilevel method by up to$70 \times$, and the proposed compression method can improve compression ratio by up to$2 \times$compared with other state-of-the-art error-bounded lossy compressors under the same level of data distortion.

Vast volumes of data are produced by today's scientific simulations and advanced instruments. These data cannot be stored and transferred efficiently because of limited I/O bandwidth, network speed, and storage capacity. Error-bounded lossy compression can be an effective method for addressing these issues: not only can it significantly reduce data size, but it can also control the data distortion based on user-defined error bounds. In practice, many scientific applications have specific requirements or constraints for lossy compression, in order to guarantee that the reconstructed data are valid for post hoc analysis. For example, some datasets contain irrelevant data that should be isolated in particular and users often have intuition regarding value ranges, geospatial regions, and other data subsets that are crucial for subsequent analysis. Existing state-of-the-art error-bounded lossy compressors, however, do not consider these constraints during compression, resulting in inferior compression ratios with respect to user's post hoc analysis, due to the fact that the data itself provides little or no value for post hoc analysis. In this work we address this issue by proposing an optimized framework that can preserve diverse constraints during the error-bounded lossy compression, e.g., cleaning the irrelevant data, efficiently preserving different precision for multiple value intervals, and allowing users to set diverse precision over both regular and irregular regions. We perform our evaluation on a supercomputer with up to 2,100 cores. Experiments with six real-world applications show that our proposed diverse constraints based error-bounded lossy compressor can obtain a higher visual quality or data fidelity on reconstructed data with the same or even higher compression ratios compared with the traditional state-of-the-art compressor SZ. Our experiments also demonstrate very good scalability in compression performance compared with the I/O throughput of the parallel file system.

In the past decades, various lossy compressors have been studied broadly due to the ever-increasing volume of data being produced by today’s scientific applications. SZ has been one of the best error-bounded lossy compressors ever raised, and it has a flexible framework that includes four adjustable steps: prediction, quantization, variable-length encoding, and lossless compression. In this paper, we improve the lossy compression performances of the SZ compression model by exploring different existing lossless compression techniques using the Squash data compression benchmark. Specifically, we first characterize the bytes outputted by the first three steps in SZ, then we investigate the best lossless compressor with different datasets and different error bounds. We perform our exploration by testing 8 widely used lossless compressors under different configurations together with SZ over five well-known scientific simulation datasets. Our experiments show that adopting the best-fit lossless compressor selected based on our analysis can improve the overall compression speed by up to 40% compared to the previous lossless compression technique used in SZ with the comparable quality of reconstructed data.

In recent years, the increasing complexity in scientific simulations and emerging demands for training heavy artificial intelligence models require massive and fast data accesses, which urges high-performance computing (HPC) platforms to equip with more advanced storage infrastructures such as solid-state disks (SSDs). While SSDs offer high-performance I/O, the reliability challenges faced by the HPC applications under the SSD-related failures remains unclear, in particular for failures resulting in data corruptions. The goal of this paper is to understand the impact of SSD-related faults on the behaviors of complex HPC applications. To this end, we propose FFIS, a FUSE-based fault injection framework that systematically introduces storage faults into the application layer to model the errors originated from SSDs. FFIS is able to plant different I/O related faults into the data returned from underlying file systems, which enables the investigation on the error resilience characteristics of the scientific file format. We demonstrate the use of FFIS with three representative real HPC applications, showing how each application reacts to the data corruptions, and provide insights on the error resilience of the widely adopted HDF5 file format for the HPC applications.

Error-bounded lossy compression is becoming an indispensable technique for the success of today’s scientific projects with vast volumes of data produced during the simulations or instrument data acquisitions. Not only can it significantly reduce data size, but it also can control the compression errors based on user-specified error bounds. Autoencoder (AE) models have been widely used in image compression, but few AE-based compression approaches support error-bounding features, which are highly required by scientific applications. To address this issue, we explore using convolutional autoencoders to improve error-bounded lossy compression for scientific data, with the following three key contributions. (1) We provide an in-depth investigation of the characteristics of various autoencoder models and develop an error-bounded autoencoder-based framework in terms of the SZ model. (2) We optimize the compression quality for main stages in our designed AE-based error-bounded compression framework, fine-tuning the block sizes and latent sizes and also optimizing the compression efficiency of latent vectors. (3) We evaluate our proposed solution using five real-world scientific datasets and comparing them with six other related works. Experiments show that our solution exhibits a very competitive compression quality from among all the compressors in our tests. In absolute terms, it can obtain a much better compression quality (100% $\sim$ 800% improvement in compression ratio with the same data distortion) compared with SZ2.1 and ZFP in cases with a high compression ratio.

Error-bounded lossy compression is a critical technique for significantly reducing scientific data volumes. With ever-emerging heterogeneous high-performance computing (HPC) architecture, GPU-accelerated error-bounded compressors (such as CUSZ and cuZFP) have been developed. However, they suffer from either low performance or low compression ratios. To this end, we propose CUSZ+ to target both high compression ratios and throughputs. We identify that data sparsity and data smoothness are key factors for high compression throughputs. Our key contributions in this work are fourfold: (1) We propose an efficient compression workflow to adaptively perform run-length encoding and/or variable-length encoding. (2) We derive Lorenzo reconstruction in decompression as multidimensional partial-sum computation and propose a fine-grained Lorenzo reconstruction algorithm for GPU architectures. (3) We carefully optimize each of CUSZ kernels by leveraging state-of-the-art CUDA parallel primitives. (4) We evaluate CU SZ+ using seven real-world HPC application datasets on V100 and A100 GPUs. Experiments show CUSZ+ improves the compression throughputs and ratios by up to 18.4× and 5.3×, respectively, over CUSZ on the tested datasets.

Lossy compression is becoming an indispensable technique for the success of today’s extreme-scale high-performance computing projects that produce vast volumes of data during scientific simulations or instrument data acquisitions. Comprehensively understanding the compression quality and performance of different lossy compressors is critical to selecting the best-fit compressors and using them properly and efficiently in practice. A few lossy compression assessment tools (e.g., Z-checker) have been developed, but none of them support the execution in a GPU environment. This is a significant gap because many recent extreme-scale applications and lossy compressors (e.g., cuSZ) can run entirely within GPUs. In this work, we develop an efficient lossy compression measuring system (called cuZ-Checker) on the GPU platform, which aims to perform the lossy compression quality and performance assessment completely within the GPU environment. Our contribution is threefold. (1) We develop a novel GPU-based lossy compression measuring framework using a computation pattern-based design approach. This approach classifies the computing-intensive metrics into three categories based on their patterns which creates large opportunities for kernel fusion and data reuse. (2) For each pattern in cuZ-Checker, we develop a CUDA kernel and provide fine-grained optimizations to boost its performance. (3) We thoroughly evaluate our cuZ-checker on a V100 GPU using four real-world scientific application datasets. Experiments show that cuZ-Checker can significantly accelerate the overall lossy compression assessment performance by $23X \sim31X$ compared with the OpenMP-based multithreading CPU performance. To the best of our knowledge, this is the first lossy compression measuring system designed for GPU devices.

Vast volumes of scientific data cannot be stored and transferred efficiently because of limited I/O bandwidth, network bandwidth, and storage capacity. Error-bounded lossy compression can be an effective method for resolving these big data issues, since not only can it significantly reduce the data size but it can also control the data distortion based on user-defined error bounds. In practice, many scientific applications have specific data fidelity requirements across different value ranges/intervals of the dataset for the lossy compression, in order to guarantee that the reconstructed data are valid for post hoc analysis. Existing state-of-the-art error-bounded lossy compressors, however, do not support multi-range based error-bounds in the lossy compression, leaving a critical gap that hampers their effective use in practice. In this work, we address this issue by proposing a multi-range based error-bounded lossy compressor based on the state-of-the-art SZ lossy compressor. Our approach allows users to set different error bounds in different value ranges for a compressoin task. We evaluate our approach on several real-world datasets and show that it can obtain a higher visual quality or data fidelity on reconstructed data with the same or even higher compression ratios achieved by SZ.

Extreme-scale cosmological simulations have been widely used by today's researchers and scientists on leadership supercomputers. A new generation of error-bounded lossy compressors has been used in workflows to reduce storage requirements and minimize the impact of throughput limitations while saving large snapshots of high-fidelity data for post-hoc analysis. In this paper, we propose to adaptively provide compression configurations to compute partitions of cosmological simulations with newly designed post-analysis aware rate-quality modeling. The contribution is fourfold: (1) We propose a novel adaptive approach to select feasible error bounds for different partitions, showing the possibility and efficiency of adaptively configuring lossy compression for each partition individually. (2) We build models to estimate the overall loss of post-analysis result due to lossy compression and to estimate compression ratio, based on the property of each partition. (3) We develop an efficient optimization guideline to determine the best-fit configuration of error bounds combination in order to maximize the compression ratio under acceptable post-analysis quality loss. (4) Our approach introduces negligible overheads for feature extraction and error-bound optimization for each partition, enabling post-analysis-aware in situ lossy compression for cosmological simulations. Experiments show that our proposed models are highly accurate and reliable. Our fine-grained adaptive configuration approach improves the compression ratio of up to 73% on the tested datasets with the same post-analysis distortion with only 1% performance overhead.

Convolutional neural networks (CNNs) are becoming increasingly deeper, wider, and non-linear because of the growing demand on prediction accuracy and analysis quality. The wide and deep CNNs, however, require a large amount of computing resources and processing time. Many previous works have studied model pruning to improve inference performance, but little work has been done for effectively reducing training cost. In this paper, we propose ClickTrain: an efficient and accurate end-to-end training and pruning framework for CNNs. Different from the existing pruning-during-training work, ClickTrain provides higher model accuracy and compression ratio via fine-grained architecture-preserving pruning. By leveraging pattern-based pruning with our proposed novel accurate weight importance estimation, dynamic pattern generation and selection, and compiler-assisted computation optimizations, ClickTrain generates highly accurate and fast pruned CNN models for direct deployment without any time overhead, compared with the baseline training. ClickTrain also reduces the end-to-end time cost of the state-of-the-art pruning-after-training method by up to 2.3x with comparable accuracy and compression ratio. Moreover, compared with the state-of-the-art pruning-during-training approach, ClickTrain provides significant improvements both accuracy and compression ratio on the tested CNN models and datasets, under similar limited training time.

Today's high-performance computing (HPC) applications are producing vast volumes of data, which are challenging to store and transfer efficiently during the execution, such that data compression is becoming a critical technique to mitigate the storage burden and data movement cost. Huffman coding is arguably the most efficient Entropy coding algorithm in information theory, such that it could be found as a fundamental step in many modern compression algorithms such as DEFLATE. On the other hand, today's HPC applications are more and more relying on the accelerators such as GPU on supercomputers, while Huffman encoding suffers from low throughput on GPUs, resulting in a significant bottleneck in the entire data processing. In this paper, we propose and implement an efficient Huffman encoding approach based on modern GPU architectures, which addresses two key challenges: (1) how to parallelize the entire Huffman encoding algorithm, including codebook construction, and (2) how to fully utilize the high memory-bandwidth feature of modern GPU architectures. The detailed contribution is fourfold. (1) We develop an efficient parallel codebook construction on GPUs that scales effectively with the number of input symbols. (2) We propose a novel reduction based encoding scheme that can efficiently merge the codewords on GPUs. (3) We optimize the overall GPU performance by leveraging the state-of-the-art CUDA APIs such as Cooperative Groups. (4) We evaluate our Huffman encoder thoroughly using six real-world application datasets on two advanced GPUs and compare with our implemented multithreaded Huffman encoder. Experiments show that our solution can improve the encoding throughput by up to 5.0× and 6.8× on NVIDIA RTX 5000 and V100, respectively, over the state-of-the-art GPU Huffman encoder, and by up to 3.3× over the multithread encoder on two 28-core Xeon Platinum 8280 CPUs.

DNNs are becoming increasingly deeper, wider, and nonlinear due to the growing demands on prediction accuracy and analysis quality. When training a DNN model, the intermediate activation data must be saved in the memory during forward propagation and then restored for backward propagation. Traditional memory saving techniques such as data recomputation and migration either suffers from a high performance overhead or is constrained by specific interconnect technology and limited bandwidth. In this paper, we propose a novel memory-driven high performance CNN training framework that leverages error-bounded lossy compression to significantly reduce the memory requirement for training in order to allow training larger neural networks. Specifically, we provide theoretical analysis and then propose an improved lossy compressor and an adaptive scheme to dynamically configure the lossy compression error-bound and adjust the training batch size to further utilize the saved memory space for additional speedup. We evaluate our design against state-of-the-art solutions with four widely-adopted CNNs and the ImangeNet dataset. Results demonstrate that our proposed framework can significantly reduce the training memory consumption by up to 13.5× and 1.8× over the baseline training and state-of-the-art framework with compression, respectively, with little or no accuracy loss. The full paper can be referred to at https://arxiv.org/abs/2011.09017.

Uncertain or probabilistic graphs have been ubiquitously used in many emerging applications. Previously CPU based techniques were proposed to use sampling but suffer from (1) low computation efficiency and large memory overhead, (2) low degree of parallelism, and (3) nonexistent general framework to effectively support programming uncertain graph applications. To tackle these challenges, we propose a general uncertain graph processing framework for multi-GPU systems, named BPGraph. Integrated with our highly-efficient path sampling method, BPGraph can support a wide range of uncertain graph algorithms' development and optimization. Extensive evaluation demonstrates a significant performance improvement from BPGraph over the state-of-the-art uncertain graph sampling techniques.

Deep neural networks (DNNs) are becoming increasingly deeper, wider, and non-linear due to the growing demands on prediction accuracy and analysis quality. Training wide and deep neural networks require large amounts of storage resources such as memory because the intermediate activation data must be saved in the memory during forward propagation and then restored for backward propagation. However, state-of-the-art accelerators such as GPUs are only equipped with very limited memory capacities due to hardware design constraints, which significantly limits the maximum batch size and hence performance speedup when training large-scale DNNs. Traditional memory saving techniques either suffer from performance overhead or are constrained by limited interconnect bandwidth or specific interconnect technology. In this paper, we propose a novel memory-efficient CNN training framework (called COMET) that leverages error-bounded lossy compression to significantly reduce the memory requirement for training in order to allow training larger models or to accelerate training. Our framework purposely adopts error-bounded lossy compression with a strict error-controlling mechanism. Specifically, we perform a theoretical analysis on the compression error propagation from the altered activation data to the gradients, and empirically investigate the impact of altered gradients over the training process. Based on these analyses, we optimize the error-bounded lossy compression and propose an adaptive error-bound control scheme for activation data compression. Experiments demonstrate that our proposed framework can significantly reduce the training memory consumption by up to 13.5X over the baseline training and 1.8X over another state-of-the-art compression-based framework, respectively, with little or no accuracy loss.

Error-bounded lossy compression is a state-of-the-art data reduction technique for HPC applications because it not only significantly reduces storage overhead but also can retain high fidelity for postanalysis. Because supercomputers and HPC applications are becoming heterogeneous using accelerator-based architectures, in particular GPUs, several development teams have recently released GPU versions of their lossy compressors. However, existing state-of-the-art GPU-based lossy compressors suffer from either low compression and decompression throughput or low compression quality. In this paper, we present an optimized GPU version, cuSZ, for one of the best error-bounded lossy compressors-SZ. To the best of our knowledge, cuSZ is the first error-bounded lossy compressor on GPUs for scientific data. Our contributions are fourfold. (1) We propose a dual-quantization scheme to entirely remove the data dependency in the prediction step of SZ such that this step can be performed very efficiently on GPUs. (2) We develop an efficient customized Huffman coding for the SZ compressor on GPUs. (3) We implement cuSZ using CUDA and optimize its performance by improving the utilization of GPU memory bandwidth. (4) We evaluate our cuSZ on five real-world HPC application datasets from the Scientific Data Reduction Benchmarks and compare it with other state-of-the-art methods on both CPUs and GPUs. Experiments show that our cuSZ improves SZ's compression throughput by up to 370.1x and 13.1x, respectively, over the production version running on single and multiple CPU cores, respectively, while getting the same quality of reconstructed data. It also improves the compression ratio by up to 3.48x on the tested data compared with another state-of-the-art GPU supported lossy compressor.

Error-bounded lossy compression is becoming more and more important to today's extreme-scale HPC applications because of the ever-increasing volume of data generated because it has been widely used in in-situ visualization, data stream intensity reduction, storage reduction, I/O performance improvement, checkpoint/restart acceleration, memory footprint reduction, etc. Although many works have optimized ratio, quality, and performance for different error-bounded lossy compressors, there is none of the existing works attempting to systematically understand the impact of lossy compression errors on HPC application due to error propagation.In this paper, we propose and develop a lossy compression fault injection tool, called LCFI. To the best of our knowledge, this is the first fault injection tool that helps both lossy compressor developers and users to systematically and comprehensively understand the impact of lossy compression errors on HPC programs. The contributions of this work are threefold: (1) We propose an efficient approach to inject lossy compression errors according to a statistical analysis of compression errors for different state-of-the-art compressors. (2) We build a fault injector which is highly applicable, customizable, easy-to-use in generating top-down comprehensive results, and demonstrate the use of LCFI. (3) We evaluate LCFI on four representative HPC benchmarks with different abstracted fault models and make several observations about error propagation and their impacts on program outputs.

Efficient error-controlled lossy compressors are becoming critical to the success of today's large-scale scientific applications because of the ever-increasing volume of data produced by the applications. In the past decade, many lossless and lossy compressors have been developed with distinct design principles for different scientific datasets in largely diverse scientific domains. In order to support researchers and users assessing and comparing compressors in a fair and convenient way, we establish a standard compression assessment benchmark - Scientific Data Reduction Benchmark (SDRBench)1. SDRBench contains a vast variety of real-world scientific datasets across different domains, summarizes several critical compression quality evaluation metrics, and integrates many state-of-the-art lossy and lossless compressors. We demonstrate evaluation results using SDRBench and summarize six valuable takeaways that are helpful to the in-depth understanding of lossy compressors.

Recurrent neural networks (RNNs) based automatic speech recognition has nowadays become promising and important on mobile devices such as smart phones. However, previous RNN compression techniques either suffer from hardware performance overhead due to irregularity or significant accuracy loss due to the preserved regularity for hardware friendliness. In this work, we propose RTMobile that leverages both a novel block-based pruning approach and compiler optimizations to accelerate RNN inference on mobile devices. Our proposed RTMobile is the first work that can achieve real-time RNN inference on mobile platforms. Experimental results demonstrate that RTMobile can significantly outperform existing RNN hardware acceleration methods in terms of both inference accuracy and time. Compared with prior work on FPGA, RTMobile using Adreno 640 embedded GPU on GRU can improve the energy-efficiency by 40× while maintaining the same inference time.

Today's extreme-scale high-performance computing (HPC) applications are producing volumes of data too large to save or transfer because of limited storage space and I/O bandwidth. Error-bounded lossy compression has been commonly known as one of the best solutions to the big science data issue, because it can significantly reduce the data volume with strictly controlled data distortion based on user requirements. In this work, we develop an adaptive parameter optimization algorithm integrated with a series of optimization strategies for SZ, a state-of-the-art prediction-based compression model. Our contribution is threefold. (1) We exploit effective strategies by using 2nd-order regression and 2nd-order Lorenzo predictors to improve the prediction accuracy significantly for SZ, thus substantially improving the overall compression quality. (2) We design an efficient approach selecting the best-fit parameter setting, by conducting a comprehensive priori compression quality analysis and exploiting an efficient online controlling mechanism. (3) We evaluate the compression quality and performance on a supercomputer with 4,096 cores, as compared with other state-of-the-art error-bounded lossy compressors. Experiments with multiple real-world HPC simulations datasets show that our solution can improve the compression ratio up to 46% compared with the second-best compressor. Moreover, the parallel I/O performance is improved by up to 40% thanks to the significant reduction of data size.

Deep neural networks (DNNs) have gained considerable attention in various real-world applications due to the strong performance on representation learning. However, a DNN needs to be trained many epochs for pursuing a higher inference accuracy, which requires storing sequential versions of DNNs and releasing the updated versions to users. As a result, large amounts of storage and network resources are required, significantly hampering DNN utilization on resource-constrained platforms (e.g., IoT, mobile phone).

To help understand our universe better, researchers and scientists currently run extreme-scale cosmology simulations on leadership supercomputers. However, such simulations can generate large amounts of scientific data, which often result in expensive costs in data associated with data movement and storage. Lossy compression techniques have become attractive because they significantly reduce data size and can maintain high data fidelity for post-analysis. In this paper, we propose to use GPU-based lossy compression for extreme-scale cosmological simulations. Our contributions are threefold: (1) we implement multiple GPU-based lossy compressors to our open-source compression benchmark and analysis framework named Foresight; (2) we use Foresight to comprehensively evaluate the practicality of using GPU-based lossy compression on two real-world extreme-scale cosmology simulations, namely HACC and Nyx, based on a series of assessment metrics; and (3) we develop a general optimization guideline on how to determine the best-fit configurations for different lossy compressors and cosmological simulations. Experiments show that GPU-based lossy compression can provide necessary accuracy on post-analysis for cosmological simulations and high compression ratio of 5 ~ 15× on the tested datasets, as well as much higher compression and decompression throughput than CPU-based compressors.

Error-bounded lossy compression is critical to the success of extreme-scale scientific research because of ever-increasing volumes of data produced by today's high-performance computing (HPC) applications. Not only can error-controlled lossy compressors significantly reduce the I/O and storage burden but they can retain high data fidelity for post analysis. Existing state-of-the-art lossy compressors, however, generally suffer from relatively low compression and decompression throughput (up to hundreds of megabytes per second on a single CPU core), which considerably restrict the adoption of lossy compression by many HPC applications especially those with a fairly high data production rate. In this paper, we propose a highly efficient lossy compression approach based on field programmable gate arrays (FPGAs) under the state-of-the-art lossy compression model SZ. Our contributions are fourfold. (1) We adopt a wavefront memory layout to alleviate the data dependency during the prediction for higher-dimensional predictors, such as the Lorenzo predictor. (2) We propose a co-design framework named waveSZ based on the wavefront memory layout and the characteristics of SZ algorithm and carefully implement it by using high-level synthesis. (3) We propose a hardware-algorithm co-optimization method to improve the performance. (4) We evaluate our proposed waveSZ on three real-world HPC simulation datasets from the Scientific Data Reduction Benchmarks and compare it with other state-of-the-art methods on both CPUs and FPGAs. Experiments show that our waveSZ can improve SZ's compression throughput by 6.9X ~ 8.7X over the production version running on a state-of-the-art CPU and improve the compression ratio and throughput by 2.1X and 5.8X on average, respectively, compared with the state-of-the-art FPGA design.

Scientific simulations in high-performance computing (HPC) environments generate vast volume of data, which may cause a severe I/O bottleneck at runtime and a huge burden on storage space for postanalysis. Unlike traditional data reduction schemes such as deduplication or lossless compression, not only can error-controlled lossy compression significantly reduce the data size but it also holds the promise to satisfy user demand on error control. Pointwise relative error bounds (i.e., compression errors depends on the data values) are widely used by many scientific applications with lossy compression since error control can adapt to the error bound in the dataset automatically. Pointwise relative-error-bounded compression is complicated and time consuming. We develop efficient precomputation-based mechanisms based on the SZ lossy compression framework. Our mechanisms can avoid costly logarithmic transformation and identify quantization factor values via a fast table lookup, greatly accelerating the relative-error-bounded compression with excellent compression ratios. In addition, we reduce traversing operations for Huffman decoding, significantly accelerating the decompression process in SZ. Experiments with eight well-known real-world scientific simulation datasets show that our solution can improve the compression and decompression rates (i.e., the speed) by about 40 and 80 p, respectively, in most of cases, making our designed lossy compression strategy the best-in-class solution in most cases.

In the past decade, we have envisioned enormous growth in the data generated by different sources, ranging from weather sensors and customer purchasing records to Internet of Things devices. Emerging data-driven technologies have been reforming our daily life for years, such as Amazon Personalize [1], which creates real-time individualized recommendations for customers according to multidimensional data analytics. It is, however, a challenging task to fully utilize and harness the potential of data, especially big data, due to Volume, Velocity, Variety, Variability and Value (5Vs) [2]. Most businesses thus choose to migrate their hardware demands to cloud providers, such as Amazon Web Service [3], which is powered by hundreds of thousands of servers. A cluster that builds up by a number of cloud servers is a basic management unit to provide shared computing resources. The typical structure of a cluster consists of managers and workers. When a job arrives at the cluster, as the first step, managers have to select a worker to host the incoming job. Traditionally, the selection process is based on the state of the workers, e.g., resource availability and specifications of jobs, e.g., labels, zones and regions. With respect to currently running jobs, we propose a progress based container placement scheme, named ProCon. When scheduling incoming containers, ProCon not only considers instant resource utilization on the workers but also takes into account the estimation of future resource usage. Through monitoring the progress of running jobs, ProCon balances the resource contentions across the cluster and reduces the completion time as well as the makespan. Specifically, extensive experiments prove that ProCon reduces completion time by up to 53.3% for a particular job and improves overall performance by 23.0%. Additionally, ProCon records an improvement of makespan for up to 37.4% when compared to the default scheduler available in Kubernetes.

In memory data analytic frameworks like Apache Spark are employed by an increasing number of diverse applications-such as machine learning, graph computation, and scientific computing, which benefit from the long-running process (e.g. executor) programming model to avoid system I/O overhead. However, existing resource allocation strategies mainly rely on the peak demand normally specified by users. Since the resource usages of long-running applications like iterative computation vary significantly over time, we find that peak-demand-based resource allocation policies lead to low cloud utilization in production environments. In this paper, we present an elastic utilization aware executor provisioning approach for iterative workloads on Apache Spark (i.e., iSpark). It can identify the causes of resource underutilization due to an inflexible resource policy, and elastically adjusts the allocated executors over time according to the real-time resource usage. In general, iterative applications require more computation resources at the beginning stage and their demands for resources diminish as more iterations are completed. iSpark aims to timely scale up or scale down the number of executors in order to fully utilize the allocated resources while taking the dominant factor into consideration. It further preempts the underutilized executors and preserves the cached intermediate data to ensure the data consistency. Testbed evaluations show that iSpark averagely improves the resource utilization of individual executors by 35.2 % compared to vanilla Spark. At the same time, it increases the cluster utilization from 32.1% to 51.3% and effectively reduces the overall job completion time by 20.8% for a set of representative iterative applications.

Because of the ever-increasing data being produced by today's high performance computing (HPC) scientific simulations, I/O performance is becoming a significant bottleneck for their executions. An efficient error-controlled lossy compressor is a promising solution to significantly reduce data writing time for scientific simulations running on supercomputers. In this paper, we explore how to optimize the data dumping performance for scientific simulation by leveraging error-bounded lossy compression techniques. The contributions of the paper are threefold. (1) We propose a novel I/O performance profiling model that can effectively represent the I/O performance with different execution scales and data sizes, and optimize the estimation accuracy of data dumping performance using least square method. (2) We develop an adaptive lossy compression framework that can select the bestfit compressor (between two leading lossy compressors SZ and ZFP) with optimized parameter settings with respect to overall data dumping performance. (3) We evaluate our adaptive lossy compression framework with up to 32k cores on a supercomputer facilitated with fast I/O systems and using real-world scientific simulation datasets. Experiments show that our solution can mostly always lead the data dumping performance to the optimal level with very accurate selection of the bestfit lossy compressor and settings. The data dumping performance can be improved by up to 27% at different scales.

Today's deep neural networks (DNNs) are becoming deeper and wider because of increasing demand on the analysis quality and more and more complex applications to resolve. The wide and deep DNNs, however, require large amounts of resources (such as memory, storage, and I/O), significantly restricting their utilization on resource-constrained platforms. Although some DNN simplification methods (such as weight quantization) have been proposed to address this issue, they suffer from either low compression ratios or high compression errors, which may introduce an expensive fine-tuning overhead (i.e., a costly retraining process for the target inference accuracy). In this paper, we propose DeepSZ: an accuracy-loss expected neural network compression framework, which involves four key steps: network pruning, error bound assessment, optimization for error bound configuration, and compressed model generation, featuring a high compression ratio and low encoding time. The contribution is threefold. (1)We develop an adaptive approach to select the feasible error bounds for each layer. (2) We build a model to estimate the overall loss of inference accuracy based on the inference accuracy degradation caused by individual decompressed layers. (3) We develop an efficient optimization algorithm to determine the best-fit configuration of error bounds in order to maximize the compression ratio under the user-set inference accuracy constraint. Experiments show that DeepSZ can compress AlexNet and VGG-16 on the ImageNet dataset by a compression ratio of 46× and 116×, respectively, and compress LeNet-300-100 and LeNet-5 on the MNIST dataset by a compression ratio of 57× and 56×, respectively, with only up to 0.3% loss of inference accuracy. Compared with other state-of-the-art methods, DeepSZ can improve the compression ratio by up to 1.43×, the DNN encoding performance by up to 4.0× with four V100 GPUs, and the decoding performance by up to 6.2×.

Linear algebra operations have been widely used in big data analytics and scientific computations. Many works have been done on optimizing linear algebra operations on GPUs with regular-shaped input. However, few works are focusing on fully utilizing GPU resources when the input is not regular-shaped. Current optimizations lack of considering fully utilizing the memory bandwidth and computing power, therefore they could only achieve sub-optimal performance. In this paper, we propose a performant tall-and-skinny matrix-matrix multiplication algorithm on GPUs - TSM2. It focuses on optimizing linear algebra operation with none regular-shaped input. We implement the proposed algorithm and test on three different Nvidia GPU micro-architectures: Kepler, Maxwell, and Pascal. Experiments show that our TSM2 speedups the computation by 1.1x - 3x, improves memory bandwidth utilization by 8% - 47.6%, and improves computing power utilization by 7% - 37.3% comparing to the current state-of-the-art works. We replace the original matrix operations in K-means and Algorithm-Bases Fault Tolerance (ABFT) with TSM2 and achieve up to 1.89x and 1.90x speed up.

Scientific simulations in high-performance computing (HPC) environments are producing vast volume of data, which may cause a severe I/O bottleneck at runtime and a huge burden on storage space for post-analysis. Unlike the traditional data reduction schemes (such as deduplication or lossless compression), not only can error-controlled lossy compression significantly reduce the data size but it can also hold the promise to satisfy user demand on error control. Point-wise relative error bounds (i.e., compression errors depends on the data values) are widely used by many scientific applications in the lossy compression, since error control can adapt to the precision in the dataset automatically. Point-wise relative error bounded compression is complicated and time consuming. In this work, we develop efficient precomputation-based mechanisms in the SZ lossy compression framework. Our mechanisms can avoid costly logarithmic transformation and identify quantization factor values via a fast table lookup, greatly accelerating the relative-error bounded compression with excellent compression ratios. In addition, our mechanisms also help reduce traversing operations for Huffman decoding, and thus significantly accelerate the decompression process in SZ. Experiments with four well-known real-world scientific simulation datasets show that our solution can improve the compression rate by about 30% and decompression rate by about 70% in most of cases, making our designed lossy compression strategy the best choice in class in most cases.

With the ever-increasing volumes of data produced by today's large-scale scientific simulations, error-bounded lossy compression techniques have become critical: not only can they significantly reduce the data size but they also can retain high data fidelity for postanalysis. In this paper, we design a strategy to improve the compression quality significantly based on an optimized, hybrid prediction model. Our contribution is fourfold. (1) We propose a novel, transform-based predictor and optimize its compression quality. (2) We significantly improve the coefficient-encoding efficiency for the data-fitting predictor. (3) We propose an adaptive framework that can select the best-fit predictor accurately for different datasets. (4) We evaluate our solution and several existing state-of-the-art lossy compressors by running real-world applications on a supercomputer with 8,192 cores. Experiments show that our adaptive compressor can improve the compression ratio by 112~165% compared with the second-best compressor. The parallel I/O performance is improved by about 100% because of the significantly reduced data size. The total I/O time is reduced by up to 60X with our compressor compared with the original I/O time.

An effective data compressor is becoming increasingly critical to today's scientific research, and many lossy compressors are developed in the context of absolute error bounds. Based on physical/chemical definitions of simulation fields or multiresolution demand, however, many scientific applications need to compress the data with a pointwise relative error bound (i.e., the smaller the data value, the smaller the compression error to tolerate). To this end, we propose two optimized lossy compression strategies under a state-of-the-art three-staged compression framework (prediction + quantization + entropy-encoding). The first strategy (called block-based strategy) splits the data set into many small blocks and computes an absolute error bound for each block, so it is particularly suitable for the data with relatively high consecutiveness in space. The second strategy (called multi-threshold-based strategy) splits the whole value range into multiple groups with exponentially increasing thresholds and performs the compression in each group separately, which is particularly suitable for the data with a relatively large value range and spiky value changes. We implement the two strategies rigorously and evaluate them comprehensively by using two scientific applications which both require lossy compression with point-wise relative error bound. Experiments show that the two strategies exhibit the best compression qualities on different types of data sets respectively. The compression ratio of our lossy compressor is higher than that of other state-of-the-art compressors by 17.2-618 percent on the climate simulation data and 30-210 percent on the N-body simulation data, with the same relative error bound and without degradation of the overall visualization effect of the entire data.

With ever-increasing volumes of scientific data produced by high-performance computing applications, significantly reducing data size is critical because of limited capacity of storage space and potential bottlenecks on I/O or networks in writing/reading or transferring data. SZ and ZFP are two leading BSD licensed open source C/C++ libraries for compressed floating-point arrays that support high throughput read and write random access. However, their performance is not consistent across different data sets and across different fields of some data sets, which raises the need for an automatic online (during compression) selection between SZ and ZFP, with minimal overhead. In this paper, the automatic selection optimizes the rate-distortion, an important statistical quality metric based on the signal-to-noise ratio. To optimize for rate-distortion, we investigate the principles of SZ and ZFP. We then propose an efficient online, low-overhead selection algorithm that predicts the compression quality accurately for two compressors in early processing stages and selects the best-fit compressor for each data field. We implement the selection algorithm into an open-source library, and we evaluate the effectiveness of our proposed solution against plain SZ and ZFP in a parallel environment with 1,024 cores. Evaluation results on three data sets representing about 100 fields show that our selection algorithm improves the compression ratio up to 70 percent with the same level of data distortion because of very accurate selection (around 99 percent) of the bestfit compressor, with little overhead (less than 7 percent in the experiments).

Today's scientific simulations require a significant reduction of the data size because of extremely large volumes of data they produce and the limitation of storage bandwidth and space. If the compression is set to reach a high compression ratio, however, the reconstructed data are often distorted too much to tolerate. In this paper, we explore a new compression strategy that can effectively control the data distortion when significantly reducing the data size. The contribution is threefold. (1) We propose an adaptive compression framework to select either our improved Lorenzo prediction method or our optimized linear regression method dynamically in different regions of the dataset. (2) We explore how to select them accurately based on the data features in each block to obtain the best compression quality. (3) We analyze the effectiveness of our solution in details using four real-world scientific datasets with 100+ fields. Evaluation results confirm that our new adaptive solution can significantly improve the rate distortion for the lossy compression with fairly high compression ratios. The compression ratio of our compressor is 1.5X~8X as high as that of two other leading lossy compressors (SZ and ZFP) with the same peak single-to-noise ratio (PSNR), in the high-compression cases. Parallel experiments with 8,192 cores and 24 TB of data shows that our solution obtains 1.86X dumping performance and 1.95X loading performance compared with the second-best lossy compressor, respectively.

Computation of two-electron repulsion integrals is the critical and the most time-consuming step in a typical parallel quantum chemistry simulation. Such calculations have massive computing and storage requirements, which scale as O(N^4) with the size of a chemical system. Compressing the integral's data and storing it on disk can avoid costly recalculation, significantly speeding the overall quantum chemistry calculations; but it requires a fast compression algorithm. To this end, we developed PaSTRI (Pattern Scaling for Two-electron Repulsion Integrals) and implemented the algorithm in the data compression package SZ. PaSTRI leverages the latent pattern features in the integral dataset and optimizes the calculation of the appropriate number of bits required for the storage of the integral. We have evaluated PaSTRI using integral datasets generated by the quantum chemistry program GAMESS. The results show an excellent 16.8 compression ratio with low overhead, while maintaining 10-10absolute precision based on user's requirement.

Because of the ever-increasing execution scale of scientific applications, how to store the extremely large volume of data efficiently is becoming a serious issue. A significant reduction of the scientific data size can effectively mitigate the I/O burden and save considerable storage space. Since lossless compressors suffer from limited compression ratios, error-controlled lossy compressors have been studied for years. Existing error-controlled lossy compressors, however, focus mainly on absolute error bounds, which cannot meet users' diverse demands such as pointwise relative error bounds. Although some of the state-of-the-art lossy compressors support pointwise relative error bound, the compression ratios are generally low because of the limitation in their designs and possible spiky data changes in local data regions. In this work, we propose a novel, efficient approach to perform compression based on the pointwise relative error bound with higher compression ratios than existing solutions provide. Our contribution is threefold. (1) We propose a novel transformation scheme that can transfer the pointwise relative-error-bounded compression problem to an absolute-error-bounded compression issue. We also analyze the practical properties of our transformation scheme both theoretically and experimentally. (2) We implement the proposed technique in two of the most popular absolute-error-bounded lossy compressors, SZ and ZFP. (3) We evaluate our solution using multiple real-world application data across different scientific domains on a supercomputer with up to 4,096 cores and 12 TB of data. Experiments show that our solution achieves over 1.38X dumping and 1.31X loading performance over the second-best lossy compressor, respectively.

Error-controlled lossy compression has been studied for years because of extremely large volumes of data being produced by today's scientific simulations. None of existing lossy compressors, however, allow users to fix the peak signal-to-noise ratio (PSNR) during compression, although PSNR has been considered as one of the most significant indicators to assess compression quality. In this paper, we propose a novel technique providing a fixed-PSNR lossy compression for scientific data sets. We implement our proposed method based on the SZ lossy compression framework and release the code as an open-source toolkit. We evaluate our fixed-PSNR compressor on three realworld high-performance computing data sets. Experiments show that our solution has a high accuracy in controlling PSNR, with an average deviation of 0.1 ~ 5.0 dB on the tested data sets.

Iterative methods are commonly used approaches to solve large, sparse linear systems, which are fundamental operations for many modern scientific simulations. When the large-scale iterative methods are running with a large number of ranks in parallel, they have to checkpoint the dynamic variables periodically in case of unavoidable fail-stop errors, requiring fast I/O systems and large storage space. To this end, significantly reducing the checkpointing overhead is critical to improving the overall performance of iterative methods. Our contribution is fourfold. (1) We propose a novel lossy checkpointing scheme that can significantly improve the checkpointing performance of iterative methods by leveraging lossy compressors. (2) We formulate a lossy checkpointing performance model and derive theoretically an upper bound for the extra number of iterations caused by the distortion of data in lossy checkpoints, in order to guarantee the performance improvement under the lossy checkpointing scheme. (3) We analyze the impact of lossy checkpointing (i.e., extra number of iterations caused by lossy checkpointing files) for multiple types of iterative methods. (4) We evaluate the lossy checkpointing scheme with optimal checkpointing intervals on a high-performance computing environment with 2,048 cores, using a well-known scientific computation package PETSc and a state-of-the-art checkpoint/restart toolkit. Experiments show that our optimized lossy checkpointing scheme can significantly reduce the fault tolerance overhead for iterative methods by 23%∼70% compared with traditional checkpointing and 20%∼58% compared with lossless-compressed checkpointing, in the presence of system failures.

In situ lossy compression allowing user-controlled data loss can significantly reduce the I/O burden. For large-scale N-body simulations where only one snapshot can be compressed at a time, the lossy compression ratio is very limited because of the fairly low spatial coherence of the particle data. In this work, we assess the state-of-the-art single-snapshot lossy compression techniques of two common N-body simulation models: cosmology and molecular dynamics. We design a series of novel optimization techniques based on the two representative real-world N-body simulation codes. For molecular dynamics simulation, we propose three compression modes (i.e., best speed, best tradeoff, best compression mode) that can refine the tradeoff between the compression rate (a.k.a., speed/throughput) and ratio. For cosmology simulation, we identify that our improved SZ is the best lossy compressor with respect to both compression ratio and rate. Its compression ratio is higher than the second-best compressor by 11% with comparable compression rate. Experiments with up to 1024 cores on the Blues supercomputer at Argonne show that our proposed lossy compression method can reduce I/O time by 80% compared with writing data directly to a parallel file system and outperforms the second-best solution by 60%. Moreover, our proposed lossy compression methods have the best rate-distortion with reasonable compression errors on the tested N-body simulation data compared with state-of-the-art compressors.

Today's HPC applications are producing extremely large amounts of data, such that data storage and analysis are becoming more challenging for scientific research. In this work, we design a new error-controlled lossy compression algorithm for large-scale scientific data. Our key contribution is significantly improving the prediction hitting rate (or prediction accuracy) for each data point based on its nearby data values along multiple dimensions. We derive a series of multilayer prediction formulas and their unified formula in the context of data compression. One serious challenge is that the data prediction has to be performed based on the preceding decompressed values during the compression in order to guarantee the error bounds, which may degrade the prediction accuracy in turn. We explore the best layer for the prediction by considering the impact of compression errors on the prediction accuracy. Moreover, we propose an adaptive error-controlled quantization encoder, which can further improve the prediction hitting rate considerably. The data size can be reduced significantly after performing the variable-length encoding because of the uneven distribution produced by our quantization encoder. We evaluate the new compressor on production scientific data sets and compare it with many other state-of-the-art compressors: GZIP, FPZIP, ZFP, SZ-1.1, and ISABELA. Experiments show that our compressor is the best in class, especially with regard to compression factors (or bit-rates) and compression errors (including RMSE, NRMSE, and PSNR). Our solution is better than the second-best solution by more than a 2x increase in the compression factor and 3.8x reduction in the normalized root mean squared error on average, with reasonable error bounds and user-desired bit-rates.

This paper presents an algorithm based fault tolerance method to harden three two-sided matrix factorizations against soft errors: reduction to Hessenberg form, tridiagonal form, and bidiagonal form. These two sided factorizations are usually the prerequisites to computing eigenvalues/eigenvectors and singular value decomposition. Algorithm based fault tolerance has been shown to work on three main one-sided matrix factorizations: LU, Cholesky, and QR, but extending it to cover two sided factorizations is non-trivial because there are no obvious \textit{offline, problem} specific maintenance of checksums. We thus develop an \textit{online, algorithm} specific checksum scheme and show how to systematically adapt the two sided factorization algorithms used in LAPACK and ScaLAPACK packages to introduce the algorithm based fault tolerance.

While many algorithm-based fault tolerance (ABFT) schemes have been proposed to detect soft errors offline in the fast Fourier transform (FFT) after computation finishes, none of the existing ABFT schemes detect soft errors online before the computation finishes. This paper presents an online ABFT scheme for FFT so that soft errors can be detected online and the corrupted computation can be terminated in a much more timely manner. We also extend our scheme to tolerate both arithmetic errors and memory errors, develop strategies to reduce its fault tolerance overhead and improve its numerical stability and fault coverage, and finally incorporate it into the widely used FFTW library - one of the today's fastest FFT software implementations. Experimental results demonstrate that: (1) the proposed online ABFT scheme introduces much lower overhead than the existing offline ABFT schemes; (2) it detects errors in a much more timely manner; and (3) it also has higher numerical stability and better fault coverage.

Emerging high-performance computing platforms, with large component counts and lower power margins, are anticipated to be more susceptible to soft errors in both logic circuits and memory subsystems. We present an online algorithm-based fault tolerance (ABFT) approach to efficiently detect and recover soft errors for general iterative methods. We design a novel checksum-based encoding scheme for matrix-vector multiplication that is resilient to both arithmetic and memory errors. Our design decouples the checksum updating process from the actual computation, and allows adaptive checksum overhead control. Building on this new encoding mechanism, we propose two online ABFT designs that can effectively recover from errors when combined with a checkpoint/rollback scheme. These designs are capable of addressing scenarios under different error rates. Our ABFT approaches apply to a wide range of iterative solvers that primarily rely on matrix-vector multiplication and vector linear operations. We evaluate our designs through comprehensive analytical and empirical analysis. Experimental evaluation on the Stampede supercomputer demonstrates the low performance overheads incurred by our two ABFT schemes for preconditioned CG (0.4% and 2.2%) and preconditioned BiCGSTAB (1.0% and 4.0%) for the largest SPD matrix from UFL Sparse Matrix Collection. The evaluation also demonstrates the flexibility and effectiveness of our proposed designs for detecting and recovering various types of soft errors in general iterative methods.

Algorithm based fault tolerance (ABFT) attracts renewed interest for its extremely low overhead and good scalability. However the fault model used to design ABFT has been either abstract, simplistic, or both, leaving a gap between what occurs at the architecture level and what the algorithm expects. As the fault model is the deciding factor in choosing an effective checksum scheme, the resulting ABFT techniques have seen limited impact in practice. In this paper we seek to close the gap by directly using a comprehensive architectural fault model and devise a comprehensive ABFT scheme that can tolerate multiple architectural faults of various kinds. We implement the new ABFT scheme into high performance linpack (HPL) to demonstrate the feasibility in large scale high performance benchmark. We conduct architectural fault injection experiments and large scale experiments to empirically validate its fault tolerance and demonstrate the overhead of error handling, respectively.

While many linear algebra libraries have been developed to optimize their performance, no linear algebra library considers their energy efficiency at the library design time. In this paper, we present GreenLA - an energy efficient linear algebra software package that leverages linear algebra algorithmic characteristics to maximize energy savings with negligible overhead. GreenLA is (1) energy efficient: it saves up to several times more energy than the best existing energy saving approaches that do not modify library source codes; (2) high performance: its performance is comparable to the highly optimized linear algebra library MAGMA; and (3) transparent to applications: with the same programming interface, existing MAGMA users do not need to modify their source codes to benefit from GreenLA. Experimental results demonstrate that GreenLA is able to save up to three times more energy than the best existing energy saving approaches while delivering similar performance compared to the state-of-the-art linear algebra library MAGMA.

As the size and complexity of high performance computers increase, more soft errors will be encountered during computations. Algorithm-Based Fault Tolerance (ABFT) has been proved to be a highly efficient technique to detect soft errors in dense linear algebra operations including matrix multiplication, Cholesky and LU factorization. While ABFT can also be applied to a iterative sparse linear algebra algorithm via applying it to every individual matrix-vector multiplication in the algorithm, it often introduces considerable overhead. In this paper, we propose novel extensions to ABFT to not only reduce the overhead but also protect computations that can not be protected by existing ABFT. Instead of maintaining checksums in every individual matrix-vector multiplication, we modified the algorithms so that checksums established at the beginning of the algorithms can be maintained at every iterations throughout the algorithms. Because soft errors in most iterative sparse linear algebra algorithms will propagate from one iteration to another, we do not have to verify the correctness of the checksums at each iteration to detect errors. By reducing the frequency of verification, the fault tolerance overhead can be greatly reduced. Experimental results demonstrate that, when used with local diskless checkpoints together, our approach introduces much less overhead than the existing ABFT techniques.