Real-time scientific workflows operate on continuous data streams and must produce timely, high-quality results despite executing on complex, failure-prone infrastructure. Hardware faults, network disruptions, and performance anomalies caused by resource contention or system heterogeneity can severely degrade performance and violate real-time constraints. We focus on strengthening the resilience of the producer–consumer streaming pattern, a fundamental building block of scientific streaming workflows. We present two complementary techniques: (i) a dynamic, asynchronous, non-blocking checkpointing mechanism that preserves progress without interrupting computation, and (ii) a progress-aware load redistribution strategy that detects slow workers and proactively rebalances tasks. Together, these mechanisms maintain forward progress and balanced execution even in highly error-prone environments. Experimental results show that our approach reduces the impact of failures and performance anomalies by up to 6 ×, while introducing less than 1% overhead in failure-free execution.

AI-driven scientific discovery has emerged as a transformative fifth paradigm in research, with agentic AI playing an increasingly prominent role across scientific domains. Agentic AI can enable collaborative AI-human or even fully autonomous decision-making, but it also introduces significant reliability challenges due to the dynamic and evolutionary nature of the AI agents. Specifically, foundation model-powered agents are prone to generating hallucinated, misleading, or adversarial outputs that can propagate silently through workflows and corrupt downstream results. In this paper we present a conceptual framework for a unified approach that integrates agentic workflow-level instrumentation and agent-level safeguards to enhance the reliability of the wider system, particularly critical in science. Embedding these mechanisms into a provenance-augmented infrastructure enables early detection, containment, and recovery from erroneous behavior, ultimately enhancing reliability and reproducibility in AI-assisted scientific workflows.

This paper discusses the key challenges and future research directions for privacy-preserving federated learning (PPFL), with a focus on its application to large-scale scientific artificial intelligence models, in particular, foundation models (FMs). PPFL enables collaborative model training across distributed datasets while preserving privacy—an important collaborative approach for science. We discuss the need for efficient and scalable algorithms to address the increasing complexity of FMs, particularly when dealing with heterogeneous clients. In addition, we underscore the need for developing advance privacy-preserving techniques, such as differential privacy, to balance privacy and utility in large FMs emphasizing fairness and incentive mechanisms to ensure equitable participation among heterogeneous clients. Finally, we emphasize the need for a robust software stack supporting scalable and secure PPFL deployments across multiple high-performance computing facilities. We envision that PPFL would play a crucial role to advance scientific discovery and enable large-scale, privacy-aware collaborations across science domains.

The need for real-time processing to enable automated decision making and experimental steering has driven a shift from high-performance computing workflows on a centralized system to a distributed approach that integrates remote data sources, edge devices, and diverse compute facilities. Under this paradigm, data can be processed close to the source where it is generated, thus reducing latency and bandwidth usage. System resilience is thus a key challenge, requiring distributed workflows to survive component failures and to meet stringent quality-of-service requirements, which results in the need to mitigate anomalies such as congestion and low availability of resources. To address these challenges, we propose Diaspora, a unified resilience framework that is inspired by event-driven communication patterns used in public clouds. Specifically, we propose an event fabric that extends across sites, facilities, and computations to provide timely, reliable, and accurate information about data, application, and resource status. On top of the event fabric, we build resilience-enabling services that combine QoS-aware data streaming, resilient data views, resilient compute and data resources, and anomaly detection and prediction, all of which collectively enhance workflow resilience for these scientific cases.

Distributed data storage services tailored to specific applications have grown popular in the high-performance computing (HPC) community as a way to address I/O and storage challenges. These services offer a variety of specific interfaces, semantics, and data representations. They also expose many tuning parameters, making it difficult for their users to find the best configuration for a given workload and platform. To address this issue, we develop a novel variational-autoencoder-guided asynchronous Bayesian optimization method to tune HPC storage service parameters. Our approach uses transfer learning to leverage prior tuning results and use a dynamically updated surrogate model to explore the large parameter search space in a systematic way. We implement our approach within the DeepHyper open-source framework, and apply it to the autotuning of a high-energy physics workflow on Argonne's Theta supercomputer. We show that our transfer-learning approach enables a more than 40 x search speedup over random search, compared with a 2.5 x to 10 x speedup when not using transfer learning. Additionally, we show that our approach is on par with state-of-the-art autotuning frameworks in speed and outperforms them in resource utilization and parallelization capabilities.

In situ analysis and visualization have grown increasingly popular for enabling direct access to data from high-performance computing (HPC) simulations. As a simulation progresses and interesting physical phenomena emerge, however, the data produced may become increasingly complex, and users may need to dynamically change the type and scale of in situ analysis tasks being carried out and consequently adapt the amount of resources allocated to such tasks. To date, none of the production in situ analysis frameworks offer such an elasticity feature, and for good reason: the assumption that the number of processes could vary during run time would force developers to rethink software and algorithms at every level of the in situ analysis stack. In this paper we present Colza, a data staging service with elastic in situ visualization capabilities. Colza relies on the widely used ParaView Catalyst in situ visualization framework and enables elasticity by replacing MPI with a custom collective communication library based on the Mochi suite of libraries. To the best of our knowledge, this work is the first to enable elastic in situ visualization capabilities for HPC applications on top of existing production analysis tools.

High-performance computing (HPC) software is evolving to support an increasingly diverse set of applications and heterogeneous hardware architectures. As part of this evolution, the construction of scientific software has shifted from a traditional monolithic message passing interface executable model to a coupled, services-style model in which simulations run alongside a host of distributed HPC data services within the same batch job allocation. Microservices have emerged as a powerful new way to build these distributed data services through a composition model. However, performance analysis of composed microservices is a daunting challenge. It requires collecting, monitoring, aggre-gating, and exporting performance data from multiple sources. To be effective, the design of such a monitoring solution must allow for seamless integration into HPC applications and distributed services alike, be scalable, operate with a low overhead, and take advantage of the HPC platform. We propose SYMBIOMON, a monitoring service that is built by composing high-performance microservices. We describe its design and implementation within the context of the Mochi framework. SYMBIOMON combines a time-series data model with existing Mochi data services to collect, aggregate, and export performance metrics in a distributed manner. SYMBIOMON enables seamless, low-overhead monitoring and analysis of data services and HPC applications alike. Using HEPnOS, a production-quality Mochi data service, we demonstrate the use of SYMBIOMON to identify better service configurations.

In-situ processing addresses the gap between speeds of computing and I/O capabilities by processing data close to the data source, i.e., on the same system as the data source (e.g., a simulation). However, the effective implementation of in-situ processing workflows requires the optimization of several design parameters such as where on the system workflow data analysis/visualization (ana/vis) as placed and how execution as well as the interaction and data exchanges between ana/vis are coordinated. For example, in the case of hybrid in-situ processing, interacting ana/vis may be tightly or loosely coupled depending on their placement, and this can lead to very different performance and scalability. A key challenge is deciding the most appropriate ana/vis placement, which depends on dynamic applications, workflow, and system characteristics that might change at runtime. In this paper, we present a framework to support online adaptive data analysis placement during the execution of an in-situ workflow. Specifically, the paper presents a model and architecture, and explores several data analysis placement strategies. Evaluation results show that dynamically choosing appropriate data analysis placement strategies can balance the benefits and overhead of different data analysis placement patterns to reduce in-situ processing time.

Microservices are a powerful new way of building, customizing, and deploying distributed services owing to their flexibility and maintainability. Several large-scale distributed platforms have emerged to serve the growing needs of data-centric workloads and services in commercial computing. Concurrently, high-performance computing (HPC) systems and software are rapidly evolving to meet the demands of diversified applications and heterogeneity. The interplay of hardware factors, software configuration parameters, and the flexibility offered with a microservice architecture makes it nontrivial to estimate the optimal service instantiation for a given application workload. Further, this problem is exacerbated when considering that these services operate in a dynamic and heterogeneous HPC environment. An optimally integrated service can be vastly more performant than a haphazardly integrated one. Existing performance tools for HPC either fail to understand the request-response model of communication inherent to microservices or they operate within a narrow scope, limiting the insight that can be gleaned from employing them in isolation.We propose a methodology for integrated performance analysis of HPC microservices frameworks and applications called SYMBIOSYS. We describe its design and implementation within the context of the Mochi framework. This integration is achieved by combining distributed callpath profiling and tracing with a performance data exchange strategy that collects fine-grained, low-level metrics from the RPC communication library and network layers. The result is a portable, low-overhead performance analysis setup that provides a holistic profile of the dependencies among microservices and how they interact with the Mochi RPC software stack. Using HEPnOS, a production-quality Mochi data service, we demonstrate the low-overhead operation of SYMBIOSYS at scale and use it to identify the root causes of poorly performing service configurations.

User-space HPC data services are emerging as an appealing alternative to traditional parallel file systems, because of their ability to be tailored to application needs while eliminating unnecessary overheads incurred by POSIX compliance. The High Energy Physics (HEP) community is progressively turning towards such services to enable high-throughput accesses under heavy concurrency to billions of event data produced by instruments and consumed by subsequent analysis workflows. Such services would benefit from the possibility to be rescaled up and down to adapt to changing workloads, as experimental campaigns progress , in order to optimize resource usage. This paper formalizes rescaling a distributed storage system as a multi objective optimization problem considering three criteria: load balance, data balance, and duration of the rescaling operation. We propose a heuristic for rapidly finding a good approximate solution, while allowing users to weight the criteria as needed. The heuristic is evaluated with Pufferscale, a new rescaling manager for microservice-based distributed storage systems. To validate our approach in a real-world ecosystem, we showcase the use of Pufferscale as a means to enable storage malleability in the HEPnOS storage system for HEP applications.

In the age of big data, deep learning has emerged as a powerful tool to extract insight and exploit its value, both in industry and scientific applications. One common pattern emerging in such applications is frequent checkpointing of the state of the learning model during training, needed in a variety of scenarios: analysis of intermediate states to explain features and correlations with training data, exploration strategies involving alternative models that share a common ancestor, knowledge transfer, resilience, etc. However, with increasing size of the learning models and popularity of distributed data-parallel training approaches, simple checkpointing techniques used so far face several limitations: low serialization performance, blocking I/O, stragglers due to the fact that only a single process is involved in checkpointing. This paper proposes a checkpointing technique specifically designed to address the aforementioned limitations, introducing efficient asynchronous techniques to hide the overhead of serialization and I/O, and distribute the load over all participating processes. Experiments with two deep learning applications (CANDLE and ResNet) on a pre-Exascale HPC platform (Theta) shows significant improvement over state-of-art, both in terms of checkpointing duration and runtime overhead.

Training modern deep neural network (DNN) models involves complex workflows triggered by model exploration, sensitivity analysis, explainability, etc. A key primitive in this context is the ability to clone a model training instance, i.e. “fork” the training process in a potentially different direction, which enables comparisons of different evolution paths using variations of training data and model parameters. However, in a quest improve the training throughput, a mix of data parallel, model parallel, pipeline parallel and layer-wise parallel approaches are making the problem of cloning highly complex. In this paper, we explore the problem of efficient cloning under such circumstances. To this end, we leverage several properties of data-parallel training and layer-wise parallelism to design DeepClone, a cloning approach based on augmenting the execution graph to gain direct access to tensors, which are then sharded and reconstructed asynchronously in order to minimize runtime overhead, standby duration, readiness duration. Compared with state-of-art approaches, DeepClone shows orders of magnitude improvement for several classes of DNN models.

As scientific workflows increasingly use extreme-scale resources, the imbalance between higher computational capabilities, generated data volumes, and available I/O bandwidth is limiting the ability to translate these scales into insights. In-situ workflows (and the in-situ approach) are leveraging storage levels close to the computation in novel ways in order to reduce the required I/O. However, to be effective, it is important that the mapping and execution of such in-situ workflows adopts a data-driven approach, enabling in-situ tasks to be executed flexibly based upon data content. This paper first explores the design space for data-driven in-situ workflows. Specifically, it presents a model that captures different factors that influence the mapping, execution, and performance of data-driven in-situ workflows and experimentally studies the impact of different mapping decisions and execution patterns. The paper then presents the design, implementation, and experimental evaluation of a data-driven in-situ workflow execution framework that leverages in-memory distributed data management and user-defined task-triggers to enable efficient and scalable in-situ workflow execution.

Efficient resource utilization is a major concern for large-scale computer platforms. One method used to lower energy consumption and operational cost is to reduce the amount of idle resources. This can be achieved by using malleability, namely, the possibility for resource managers to dynamically increase or decrease the amount of resources of jobs while they are running. Decommissioning (i.e., removing from the cluster) the idle nodes as soon as possible allows the resource manager to quickly reallocate those nodes to other jobs. Challenges appear when such nodes host part of a distributed storage system. Such storage systems may need to transfer large amounts of data before releasing the nodes, in order to ensure data availability and a certain level of fault tolerance. In this paper, we model and evaluate the performance of the decommission operation when relaxing the level of fault tolerance (i.e., the number of replicas) during this operation. Intuitively, this is expected to reduce the amount of data transfers needed before nodes are released, and thus allow nodes to be returned to the resource manager faster. We quantify theoretically how much time and resources are saved by such a fast decommission strategy compared with a standard decommission that does not temporarily reduce the fault-tolerance level. We establish lower bounds for the duration of the different phases of a fast decommission. We use the lower bounds to estimate when fast decommission would be useful to reduce the usage of core-hours and when not. We implement a prototype for fast decommission and experimentally validate the lower bounds on the duration of the operation and confirm in practice our theoretical findings.

Two-tiered direct network topologies such as Dragonflies have been proposed for future post-petascale and exascale machines, since they provide a high-radix, low-diameter, fast interconnection network. Such topologies call for redesigningMPI collective communication algorithms in order to attain the best performance. Yet as increasingly more applications share a machine, it is not clear how these topology-aware algorithms will react to interference with concurrent jobs accessing the same network. In this paper, we study three topology-aware broadcast algorithms, including one designed by ourselves. We evaluate their performance through event-driven simulation for small-and large-sized broadcasts (in terms of both data size and number of processes). We study the effect of different routing mechanisms on the topology-aware collective algorithms, as well as their sensitivity to network contention with other jobs. Our results show that while topology-aware algorithms dramatically reduce link utilization, their advantage in terms of latency is more limited.

While many parallel visualization tools now provide in situ visualization capabilities, the trend has been to feed such tools with large amounts of unprocessed output data and let them render everything at the highest possible resolution. This leads to an increased run time of simulations that still have to complete within a fixed-length job allocation. In this paper, we tackle the challenge of enabling in situ visualization under performance constraints. Our approach shuffles data across processes according to its content and filters out part of it in order to feed a visualization pipeline with only a reorganized subset of the data produced by the simulation. Our framework leverages fast, generic evaluation procedures to score blocks of data, using information theory, statistics, and linear algebra. It monitors its own performance and adapts dynamically to achieve appropriate visual fidelity within predefined performance constraints. Experiments on the Blue Waters supercomputer with the CM1 simulation show that our approach enables a 5x speedup with respect to the initial visualization pipeline and is able to meet performance constraints.

Concurrent accesses to the shared storage resources in current HPC machines lead to severe performance degradation caused by I/O contention. In this study, we identify some key challenges to efficiently handling interleaved data accesses, and we propose a system-wide solution to optimize global performance. We implemented and tested several I/O scheduling policies, including prioritizing specific applications by leveraging burst buffers to defer the conflicting accesses from another application and/or directing the requests to different storage servers inside the parallel file system infrastructure. The results show that we mitigate the negative effects of interference and optimize the performance up to 2x depending on the selected I/O policy.

As we move toward the exascale era, performance variability in HPC systems remains a challenge. I/O interference, a major cause of this variability, is becoming more important every day with the growing number of concurrent applications that share larger machines. Earlier research efforts on mitigating I/O interference focus on a single potential cause of interference (e.g., the network). Yet the root causes of I/O interference can be diverse. In this work, we conduct an extensive experimental campaign to explore the various root causes of I/O interference in HPC storage systems. We use microbenchmarks on the Grid'5000 testbed to evaluate how the applications' access pattern, the network components, the file system's configuration, and the backend storage devices influence I/O interference. Our studies reveal that in many situations interference is a result of bad flow control in the I/O path, rather than being caused by some single bottleneck in one of its components. We further show that interference-free behavior is not necessarily a sign of optimal performance. To the best of our knowledge, our work provides the first deep insight into the role of each of the potential root causes of interference and their interplay. Our findings can help developers and platform owners improve I/O performance and motivate further research addressing the problem across all components of the I/O stack.

The increasing gap between the computation performance of post-petascale machines and the performance of their I/O subsystem has motivated many I/O optimizations including prefetching, caching, and scheduling. In order to further improve these techniques, modeling and predicting spatial and temporal I/O patterns of HPC applications as they run has become crucial. In this paper we present Omnisc'IO, an approach that builds a grammar-based model of the I/O behavior of HPC applications and uses it to predict when future I/O operations will occur, and where and how much data will be accessed. To infer grammars, Omnisc'IO is based on StarSequitur, a novel algorithm extending Nevill-Manning's Sequitur algorithm. Omnisc'IO is transparently integrated into the POSIX and MPI I/O stacks and does not require any modification in applications or higher-level I/O libraries. It works without any prior knowledge of the application and converges to accurate predictions of any N future I/O operations within a couple of iterations. Its implementation is efficient in both computation time and memory footprint.

Unmatched computation and storage performance in new HPC systems have led to a plethora of I/O optimizations ranging from application-side collective I/O to network and disk-level request scheduling on the file system side. As we deal with ever larger machines, the interference produced by multiple applications accessing a shared parallel file system in a concurrent manner becomes a major problem. Interference often breaks single-application I/O optimizations, dramatically degrading application I/O performance and, as a result, lowering machine wide efficiency. This paper focuses on CALCioM, a framework that aims to mitigate I/O interference through the dynamic selection of appropriate scheduling policies. CALCioM allows several applications running on a supercomputer to communicate and coordinate their I/O strategy in order to avoid interfering with one another. In this work, we examine four I/O strategies that can be accommodated in this framework: serializing, interrupting, interfering and coordinating. Experiments on Argonne's BG/P Surveyor machine and on several clusters of the French Grid'5000 show how CALCioM can be used to efficiently and transparently improve the scheduling strategy between two otherwise interfering applications, given specified metrics of machine wide efficiency.

The increasing gap between the computation performance of post-petascale machines and the performance of their I/O subsystem has motivated many I/O optimizations including prefetching, caching, and scheduling techniques. In order to further improve these techniques, modeling and predicting spatial and temporal I/O patterns of HPC applications as they run has became crucial. In this paper we present Omnisc'IO, an approach that builds a grammar-based model of the I/O behavior of HPC applications and uses it to predict when future I/O operations will occur, and where and how much data will be accessed. Omnisc'IO is transparently integrated into the POSIX and MPI I/O stacks and does not require any modification in applications or higher level I/O libraries. It works without any prior knowledge of the application and converges to accurate predictions within a couple of iterations only. Its implementation is efficient in both computation time and memory footprint.

With exascale computing on the horizon, the performance variability of I/O systems represents a key challenge in sustaining high performance. In many HPC applications, I/O is concurrently performed by all processes, which leads to I/O bursts. This causes resource contention and substantial variability of I/O performance, which significantly impacts the overall application performance and, most importantly, its predictability over time. In this paper, we propose a new approach to I/O, called Damaris, which leverages dedicated I/O cores on each multicore SMP node, along with the use of shared-memory, to efficiently perform asynchronous data processing and I/O in order to hide this variability. We evaluate our approach on three different platforms including the Kraken Cray XT5 supercomputer (ranked 11th in Top500), with the CM1 atmospheric model, one of the target HPC applications for the Blue Waters postpetascale supercomputer project. By overlapping I/O with computation and by gathering data into large files while avoiding synchronization between cores, our solution brings several benefits: 1) it fully hides jitter as well as all I/O-related costs, which makes simulation performance predictable, 2) it increases the sustained write throughput by a factor of 15 compared to standard approaches, 3) it allows almost perfect scalability of the simulation up to over 9,000 cores, as opposed to state-of-the-art approaches which fail to scale, 4) it enables a 600% compression ratio without any additional overhead, leading to a major reduction of storage requirements.

As we enter the post-petascale era, scientific applications running on large-scale platforms generate increasingly larger amounts of data for checkpointing or offline visualization, which puts current storage systems under heavy pressure. Unfortunately, I/O scalability rapidly fades behind the increasing computation power available, and thereby reduced the overall application performance scalability. We consider the common case of large-scale simulations who alternate between computation phases and I/O phases. Two main approaches have been used to handle these I/O phases: 1) each process writes an individual file, leading to a very large number of files from which it is hard to retrieve scientific insights; 2) processes synchronize and use collective I/O to write to the same shared file. In both cases, because of mandatory communications betweens processes during the computation phase, all processes enter the I/O phase at the same time, which leads to huge access contention and extreme performance variability.

Hadoop is a software framework supporting the Map-Reduce programming model. It relies on the Hadoop Distributed File System (HDFS) as its primary storage system. The efficiency of HDFS is crucial for the performance of Map-Reduce applications. We substitute the original HDFS layer of Hadoop with a new, concurrency-optimized data storage layer based on the BlobSeer data management service. Thereby, the efficiency of Hadoop is significantly improved for data-intensive Map-Reduce applications, which naturally exhibit a high degree of data access concurrency. Moreover, BlobSeer's features (built-in versioning, its support for concurrent append operations) open the possibility for Hadoop to further extend its functionalities. We report on extensive experiments conducted on the Grid'5000 testbed. The results illustrate the benefits of our approach over the original HDFS-based implementation of Hadoop.