SZ3 (0.07MB, PSNR = 29.2) (c)NeurLZ (0.07MB, PSNR = 39.1)

Distributed training often suffers from high communication overhead due to large-scale gradient synchronization. Although gradient compression—particularly at 4-bit or even lower precision—significantly reduces transfer volume, it typically results in sacrifice in precision and degradation of the final model accuracy. In this work, we introduce DUO, a distributed training framework designed to mitigate accuracy degradation incurred by gradient compression without involving additional overhead. DUO achieves this by inserting an additional high-precision gradient synchronization step into a previously computation-only phase, so that its communication is fully hidden by computation. We provide a comprehensive theoretical proof of convergence for DUO and validate its effectiveness through extensive pre-training experiments on GPT models. Our results indicate that DUO effectively restores accuracy when using 4-bit gradient compression, achieving performance comparable to uncompressed training. Remarkably, DUO maintains minimal accuracy degradation even under extreme compression scenarios, including 1-bit gradients or complete omission of the low-precision gradient communication step (0-bit transmission).

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.

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.

Error-bounded lossy compression is one of the most efficient solutions to reduce the volume of scientific data. For lossy compression, progressive decompression and random-access decompression are critical features that enable on-demand data access and flexible analysis workflows. However, these features can severely degrade compression quality and speed. To address these limitations, we propose a novel streaming compression framework that supports both progressive decompression and random-access decompression while maintaining high compression quality and speed. Our contributions are three-fold: (1) we design the first compression framework that simultaneously enables both progressive decompression and random-access decompression; (2) we introduce a hierarchical partitioning strategy to enable both streaming features, along with a hierarchical prediction mechanism that mitigates the impact of partitioning and achieves high compression quality—even comparable to state-of-the-art (SOTA) non-streaming compressor SZ3; and (3) our framework delivers high compression and decompression speed, up to 6.7 × faster than SZ3.

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.

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.

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.

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. 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.

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.