After all these years and all these other shared memory programming frameworks, OpenMP is still the most popular one. However, its greater levels of non-deterministic execution makes debugging and testing more challenging. The ability to record and deterministically replay the program execution is key to address this challenge. However, scalably replaying OpenMP programs is still an unresolved problem. In this paper, we propose two novel techniques that use Distributed Clock (DC) and Distributed Epoch (DE) recording schemes to eliminate excessive thread synchronization for OpenMP record and replay. Our evaluation on representative HPC applications with ReOMP, which we used to realize DC and DE recording, shows that our approach is 2-5x more efficient than traditional approaches that synchronize on every shared-memory access. Furthermore, we demonstrate that our approach can be easily combined with MPI-level replay tools to replay non-trivial MPI+OpenMP applications. We achieve this by integrating ReOMP into ReMPI, an existing scalable MPI record-and-replay tool, with only a small MPI-scale-independent runtime overhead.

The current era of exascale supercomputing and the emergence of a computing continuum present several significant resource management challenges. These include, but are not limited to, management of complex and high-throughput scientific workflows, diverse resources such as power and network bandwidth, extremely heterogenous and dynamic systems, elasticity in user jobs, and disaggregated and converged environments. The resource models that underpin today’s job scheduling frameworks reflect the node- (or core-) centric system architectures prevalent when the frameworks were designed. Consequently, they are not suited to capturing resource relationships or dynamism. The inability to express relationships or dynamism limits their applicability to the emerging multifaceted challenges in high-performance computing (HPC) and Cloud, as well as other converged environments. In this paper, we propose a scalable graph-based resource model to overcome these challenges, which allows for representation of complex, changing resource relationships and multiple containment hierarchies. We implement this model, Fluxion, in a production-quality scheduling framework, and evaluate its performance. Additionally, we present emerging and advanced scheduling use cases that are enabled by our model.

As High Performance Computing (HPC) workflows increase in complexity, their designers seek to enable automation and flexibility offered by cloud technologies. Container orchestration through Kubernetes enables highly desirable capabilities but does not satisfy the performance demands of HPC. Kubernetes tools that automate the lifecycle of Message Passing Interface (MPI)-based applications do not scale, and the Kubernetes scheduler does not provide crucial scheduling capabilities. In this work, we detail our efforts to port CORAL-2 benchmark codes to Kubernetes on IBM Cloud and AWS EKS. We describe contributions to the MPI Operator to achieve 3,000-rank scale, a two-orders-of-magnitude improvement to state of the art. We discuss enhancements to Fluence, our scheduler plugin for Kubernetes based on the next-generation, cloud-ready Flux framework. Finally, we compare the placement decisions of Fluence with those of the Kubernetes scheduler and demonstrate that Fluence allows simulated scientific workflows to achieve up to 3× higher performance.

Workflows in High-Performance Computing (HPC) are rapidly changing towards more complex and large-scale workflows. In particular, high-throughput and ensemble workflows are becoming increasingly common. These workflows impose significant burden on current HPC scheduling systems which typically use slow, centralized schedulers. Generalized hierarchical scheduling (GHS) is a potential solution to face modern workflows but is not widely adopted in HPC yet. One difficulty hindering widespread adoption is the lack of performance models to configure and fit application requirements. The few existing models are often built on stick assumptions that can substantially reduce the analysis realism. In this paper, we reproduce the analysis and improve the realism of a state-of-the-art model presented in "An Analytical Performance Model of Generalized Hierarchical Scheduling" [1] by Herbein and co-authors. Specifically, we first reproduce four key analysis studies in the original paper and then expand the model by removing key assumptions, one at a time. In doing so, we extend the realism of the original model. We empirically validate our extended model using three different scenarios and discuss the observed accuracy.

Exploiting task parallelism is getting increasingly difficult for diverse and complex scientific workflows running on High Performance Computing (HPC) systems. In this paper, we argue that the difficulty rises from a void in the spectrum of existing data-transfer models for resolving inter-task data dependence within a workflow and propose a novel model to fill that gap: Ubique. The Ubique model combines the best from in-transit and in situ models in order for loosely coupled producer and consumer tasks to run concurrently and to resolve their data dependencies efficiently with little or no modifications to their codes, striking a balance between transparent optimization, productivity, and performance. Our preliminary evaluation suggests that Ubique can significantly outperform the parallel file system (PFS)-based model while offering automatic data transfer and synchronization which are the features lacking in many traditional models. It also identifies the performance characteristics of its key depending subsystems, which must be understood for further broadening its benefits.

High performance computing (HPC) and cloud technologies are increasingly coupled to accelerate the convergence of traditional HPC with new simulation, data analysis, machine-learning, and artificial intelligence approaches. While the HPC+cloud paradigm, or converged computing, is ushering in new scientific discoveries with unprecedented levels of workflow automation, several key mismatches between HPC and cloud technologies still preclude this paradigm from realizing its full potential. In this paper, we present a joint effort between IBM Research, Lawrence Livermore National Laboratory (LLNL), and Red Hat to address the mismatches and to bring full HPC scheduling awareness into Kubernetes, the de facto container orchestrator for cloud-native applications, which is being increasingly adopted as a key converged-computing enabler. We found Kubernetes lacking of interfaces to enable the full spectrum of converged-computing use cases in the following three areas: (A) an interface to enable HPC batch-job scheduling (e.g., locality-aware node selection), (B) an interface to enable HPC workloads or task-level scheduling, and (C) a resource co-management interface to allow HPC resource managers and Kubernetes to co-manage a resource set. We detail our methodology and present our results, whereby the advanced graph-based scheduler Fluxion – part of the open-source Flux scheduling framework – is integrated as a Kubernetes scheduler plug-in, KubeFlux. Our initial performance study shows that KubeFlux exhibits similar performance (up to measurement precision) to the default scheduler, despite KubeFlux’s considerably more sophisticated scheduling capabilities.

The importance of workflows is highlighted by the fact that they have underpinned some of the most significant discoveries of the past decades. Many of these workflows have significant computational, storage, and communication demands, and thus must execute on a range of large-scale computer systems, from local clusters to public clouds and upcoming exascale HPC platforms. Historically, infrastructures for workflow execution consisted of complex, integrated systems, developed in-house by workflow practitioners with strong dependencies on a range of legacy technologies. Due to the increasing need to support workflows, dedicated workflow systems were developed to provide abstractions for creating, executing, and adapting workflows conveniently and efficiently while ensuring portability. While these efforts are all worthwhile individually, there are now hundreds of independent workflow systems. The resulting workflow system technology landscape is fragmented, which may present significant barriers for future workflow users due to many seemingly comparable, yet usually mutually incompatible, systems that exist. In order to tackle some of these challenges, the DOE-funded ExaWorks and NSF-funded WorkflowsRI projects have organized in 2021 a series of events entitled the"Workflows Community Summit". The third edition of the ``Workflows Community Summit"explored workflows challenges and opportunities from the perspective of computing centers and facilities. The third summit brought together a small group of facilities representatives with the aim to understand how workflows are currently being used at each facility, how facilities would like to interact with workflow developers and users, how workflows fit with facility roadmaps, and what opportunities there are for tighter integration between facilities and workflows. More information at: https://workflowsri.org/summits/facilities/

Composite science workflows are gaining traction to manage the combined effects of (1) extreme hardware heterogeneity in new High Performance Computing (HPC) systems and (2) growing software complexity – effects necessitated by the convergence of traditional HPC with data sciences. Composing, analyzing, and optimizing a composite workflow remains highly challenging as the component technologies are generally developed in isolation and often feature widely varying levels of performance, scalability, and interoperability. In this paper, we propose novel workflow composition and analysis techniques to create and optimize a scalable and effective composite workflow for heterogeneous HPC centers, and define the performance space of variables that impact composite workflow performance. We present PerfFlowAspect, an Aspect Oriented Programming (AOP)-based tool to perform cross-cutting performance analysis of composite workflows and better understand the impact of key performance variables on workflows. Our solution directly addresses AOP concerns that can affect workflow performance and covers the full software lifecycle, ranging from the workflow's initial composition through performance analysis and optimization. We use our science workflow composition techniques to implement the American Heart Association Molecule Screening (AHA MoleS) workflow. Through experimentation, we demonstrate that tuning a single performance variable can improve AHA MoleS workflow performance by a factor of up to 2.45x. Our evaluation suggests that our techniques can significantly enhance the ability of a multi-disciplinary research and development team to create a high performance composite workflow.

For decades, Lawrence Livermore National Laboratory (LLNL) has been engaged in significant research, development, and support for software to enable scientific computing and, particularly, the use of high performance computing (HPC) in the NNSA mission space. In particular, the move in the mid-1990’s to simulation as a leading component of stockpile stewardship through the ASCI and the successor ASC programs, as well as the need for reliable data acquisition and control software for the National Ignition Facility, have been important drivers in building expertise in production-quality software development at LLNL. LLNL has also been a leader in the DOE SciDAC FASTMath Institute and the DOE Exascale Computing Project (ECP), both of which have striven to make scientific computing software – in particular, the enabling technologies underpinning simulation capabilities – more widely adopted and sustainable. As such, we believe that our experience can inform the broader goal of software stewardship for scientific and high-performance computing. LLNL strongly supports the formation of a new DOE ASCR program element in software stewardship and sustainment. Historically, DOE ASCR has funded applied mathematics and computer science research that has led to the development of important new capabilities and algorithms that are expressed as artifacts in research software. Such frameworks, libraries, and tools have seldom been directly funded to address the important issues of code maintenance, documentation, robustness, and community building. Software engineering and support have typically been done on the side in support of the ASCR-driven research products. DOE funding priorities have been slow to recognize that good software engineering, the kind that ensures research investments have more adoption and longevity, requires significant resources. Based upon our experiences, we have prepared this response to highlight the concerns and issues we believe to be important as DOE ASCR considers its role in scientific software stewardship. We believe that role is important and will require a significant investment of new funding to legitimately support the technologies past and future DOE ASCR investments have and will produce to facilitate their uptake and adoption in the broader scientific computing community. Following a summary of our involvement in scientific software development, the remainder our response is organized around the nine topics specifically identified in the RFI.

Exascale computers will offer transformative capabilities to combine data-driven and learning-based approaches with traditional simulation applications to accelerate scientific discovery and insight. These software combinations and integrations, however, are difficult to achieve due to challenges of coordination and deployment of heterogeneous software components on diverse and massive platforms. We present the ExaWorks project, which can address many of these challenges: ExaWorks is leading a co-design process to create a workflow software development Toolkit (SDK) consisting of a wide range of workflow management tools that can be composed and interoperate through common interfaces. We describe the initial set of tools and interfaces supported by the SDK, efforts to make them easier to apply to complex science challenges, and examples of their application to exemplar cases. Furthermore, we discuss how our project is working with the workflows community, large computing facilities as well as HPC platform vendors to sustainably address the requirements of workflows at the exascale.

An approach to reproducibility problems related to porting software across machines and compilers.

Extreme dynamic heterogeneity in high performance computing systems and the convergence of traditional HPC with new simulation, analysis, and data science approaches impose increasingly more complex requirements on resource and job management software (RJMS). However, there is a paucity of RJMS techniques that can solve key technical challenges associated with those new requirements, particularly when they are coupled. In this paper, we propose a novel dynamic and multi-level resource model approach to address three key well-known challenges individually and in combination: i.e., 1) RJMS dynamism to facilitate job and workflow adaptability, 2) integration of specialized external resources (e.g. user-centric cloud bursting), and 3) scheduling cloud orchestration framework tasks. The core idea is to combine a dynamic directed graph resource model with fully hierarchical scheduling to provide a unified solution to all three key challenges. Our empirical and analytical evaluations of the solution using our prototype extension to Fluxion, a production hierarchical graph-based scheduler, suggest that our unified solution can significantly improve flexibility, performance and scalability across all three problems in comparison to limited traditional approaches.

Scalable management of user workloads on large-scale supercomputers remains a challenge due to the tradeoff between capturing adequate detail for analysis from various data sources and minimizing overhead. Co-designed frameworks, such as IBM’s Cluster System Management (CSM), provide a unified approach and novel insights for large-scale cluster management. This paper presents a longitudinal study and detailed analysis of a first-of-its-kind dataset collected by CSM from one of the world’s fastest supercomputers – comprised of over 1.4 million jobs on heterogenous nodes over multiple years. Furthermore, by focusing on a case study for power management, we identify the strengths and limitations of current CSM power measurement techniques in production. We present a deep dive into a large-scale scientific workflow, where finer-grained monitoring reveals power fluctuations at megawatt-levels resulting from the dynamic nature of the application, which are not captured by CSM. We make our unique datasets available to the HPC community, and discuss potential mitigation strategies by analyzing both coarse-grained and fine-grained data.

Scientific workflows have been used almost universally across scientific domains, and have underpinned some of the most significant discoveries of the past several decades. Many of these workflows have high computational, storage, and/or communication demands, and thus must execute on a wide range of large-scale platforms, from large clouds to upcoming exascale high-performance computing (HPC) platforms. These executions must be managed using some software infrastructure. Due to the popularity of workflows, workflow management systems (WMSs) have been developed to provide abstractions for creating and executing workflows conveniently, efficiently, and portably. While these efforts are all worthwhile, there are now hundreds of independent WMSs, many of which are moribund. As a result, the WMS landscape is segmented and presents significant barriers to entry due to the hundreds of seemingly comparable, yet incompatible, systems that exist. As a result, many teams, small and large, still elect to build their own custom workflow solution rather than adopt, or build upon, existing WMSs. This current state of the WMS landscape negatively impacts workflow users, developers, and researchers. The "Workflows Community Summit" was held online on January 13, 2021. The overarching goal of the summit was to develop a view of the state of the art and identify crucial research challenges in the workflow community. Prior to the summit, a survey sent to stakeholders in the workflow community (including both developers of WMSs and users of workflows) helped to identify key challenges in this community that were translated into 6 broad themes for the summit, each of them being the object of a focused discussion led by a volunteer member of the community. This report documents and organizes the wealth of information provided by the participants before, during, and after the summit.

In this work, we address the problem of running HPC workloads efficiently on Kubernetes clusters. To do so, we compare the Kubernetes' default scheduler with KubeFlux, a Kubernetes plug-in scheduler built on the Flux graph-based scheduler, on a 34- node Red Hat OpenShift cluster on IBM Cloud. We detail how scheduling can affect the performance of GROMACS, a well-known HPC application, and we show that KubeFlux can improve its performance through better pod scheduling. In our tests, KubeFlux demonstrates the tendency to limit the number of subnets spanned by a job and the maximum number of pods per node, translating to a > 2x speedup over the Kubernetes default scheduler in several cases.

The advancement of machine learning techniques and the heterogeneous architectures of most current supercomputers are propelling the demand for large multiscale simulations that can automatically and autonomously couple diverse components and map them to relevant resources to solve complex problems at multiple scales. Nevertheless, despite the recent progress in workflow technologies, current capabilities are limited to coupling two scales. In the first-ever demonstration of using three scales of resolution, we present a scalable and generalizable framework that couples pairs of models using machine learning and in situ feedback. We expand upon the massively parallel Multiscale Machine -Learned Modeling Infrastructure (MuMMI), a recent, award-winning workflow, and generalize the framework beyond its original design. We discuss the challenges and learnings in executing a massive multiscale simulation campaign that utilized over 600,000 node hours on Summit and achieved more than 98% GPU occupancy for more than 83% of the time. We present innovations to enable several orders of magnitude scaling, including simultaneously coordinating 24,000 jobs, and managing several TBs of new data per day and over a billion files in total. Finally, we describe the generalizability of our framework and, with an upcoming open-source release, discuss how the presented framework may be used for new applications.

We improved the quality and reduced the time to produce machine learned models for use in small molecule antiviral design. Our globally asynchronous multi-level parallel training approach strong scales to all of Sierra with up to 97.7% efficiency. We trained a novel, character-based Wasserstein autoencoder that produces a higher quality model trained on 1.613 billion compounds in 23 minutes while the previous state of the art takes a day on 1 million compounds. Reducing training time from a day to minutes shifts the model creation bottleneck from computer job turnaround time to human innovation time. Our implementation achieves 318 PFLOPs for 17.1% of half-precision peak. We will incorporate this model into our molecular design loop enabling the generation of more diverse compounds; searching for novel, candidate antiviral drugs improves and reduces the time to synthesize compounds to be tested in the lab.

ABSTRACTThere is growing uptake of shared memory parallelism in high performance computing, and this has increased the need for data race checking during the creation of new parallel codes or parallelizing existing sequential codes. While race checking concepts and implementations have been around for many concurrency models, including tasking models such as Cilk and PThreads (e.g., the Thread Sanitizer tool), practically usable race checkers for other APIs such as OpenMP have been lagging. For example, the OpenMP parallelization of an important library (namely Hypre) was initially unsuccessful due to inexplicable nondeterminism introduced when the code was optimized, and later root-caused to a race by the then recently developed OpenMP race checker Archer [2]. The open-source Archer now enjoys significant traction within several organizations.