Publications

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

In order to achieve exascale systems, application resilience needs to be addressed. Some programming models, such as task-DAG (directed acyclic graphs) architectures, currently embed resilience features whereas traditional SPMD (single program, multiple data) and message-passing models do not. Since a large part of the community's code base follows the latter models, it is still required to take advantage of application characteristics to minimize the overheads of fault tolerance. To that end, this paper explores how recovering from hard process/node failures in a local manner is a natural approach for certain applications to obtain resilience at lower costs in faulty environments. In particular, this paper targets enabling online, semitransparent local recovery for stencil computations on current leadership-class systems as well as presents programming support and scalable runtime mechanisms. Also described and demonstrated in this paper is the effect of failure masking, which allows the effective reduction of impact on total time to solution due to multiple failures. Furthermore, we discuss, implement, and evaluate ghost region expansion and cell-to-rank remapping to increase the probability of failure masking. To conclude, this paper shows the integration of all aforementioned mechanisms with the S3D combustion simulation through an experimental demonstration (using the Titan system) of the ability to tolerate high failure rates (i.e., node failures every five seconds) with low overhead while sustaining performance at large scales. In addition, this demonstration also displays the failure masking probability increase resulting from the combination of both ghost region expansion and cell-to-rank remapping.

Obtaining multi-process hard failure resilience at the application level is a key challenge that must be overcome before the promise of exascale can be fully realized. Previous work has shown that online global recovery can dramatically reduce the overhead of failures when compared to the more traditional approach of terminating the job and restarting it from the last stored checkpoint. If online recovery is performed in a local manner further scalability is enabled, not only due to the intrinsic lower costs of recovering locally, but also due to derived effects when using some application types. In this paper we model one such effect, namely multiple failure masking, that manifests when running Stencil parallel computations on an environment when failures are recovered locally. First, the delay propagation shape of one or multiple failures recovered locally is modeled to enable several analyses of the probability of different levels of failure masking under certain Stencil application behaviors. Our results indicate that failure masking is an extremely desirable effect at scale which manifestation is more evident and beneficial as the machine size or the failure rate increase.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Extreme-scale computational science increasingly demands multiscale and multiphysics formulations. Combining software developed by independent groups is imperative: no single team has resources for all predictive science and decision support capabilities. Scientific libraries provide high-quality, reusable software components for constructing applications with improved robustness and portability. However, without coordination, many libraries cannot be easily composed. Namespace collisions, inconsistent arguments, lack of third-party software versioning, and additional difficulties make composition costly. The Extreme-scale Scientific Software Development Kit (xSDK) defines community policies to improve code quality and compatibility across independently developed packages (hypre, PETSc, SuperLU, Trilinos, and Alquimia) and provides a foundation for addressing broader issues in software interoperability, performance portability, and sustainability. The xSDK provides turnkey installation of member software and seamless combination of aggregate capabilities, and it marks first steps toward extreme-scale scientific software ecosystems from which future applications can be composed rapidly with assured quality and scalability.

Abstract not provided.

Abstract not provided.

Exascale systems promise the potential for computation atunprecedented scales and resolutions, but achieving exascale by theend of this decade presents significant challenges. A key challenge isdue to the very large number of cores and components and the resultingmean time between failures (MTBF) in the order of hours orminutes. Since the typical run times of target scientific applicationsare longer than this MTBF, fault tolerance techniques will beessential. An important class of failures that must be addressed isprocess or node failures. While checkpoint/restart (C/R) is currentlythe most widely accepted technique for addressing processor failures, coordinated, stable-storage-based global C/R might be unfeasible atexascale when the time to checkpoint exceeds the expected MTBF. This paper explores transparent recovery via implicitly coordinated, diskless, application-driven checkpointing as a way to tolerateprocess failures in MPI applications at exascale. The discussedapproach leverages User Level Failure Mitigation (ULFM), which isbeing proposed as an MPI extension to allow applications to createpolicies for tolerating process failures. Specifically, this paper demonstrates how different implementations ofapplication-driven in-memory checkpoint storage and recovery comparein terms of performance and scalability. We also experimentally evaluate the effectiveness and scalability ofthe Fenix online global recovery framework on a production system-the Titan Cray XK7 at ORNL-and demonstrate the ability of Fenix totolerate dynamically injected failures using the execution of fourbenchmarks and mini-applications with different behaviors.

Exascale studies project reliability challenges for future HPC systems. We present the Global View Resilience (GVR) system, a library for portable resilience. GVR begins with a subset of the Global Arrays interface, and adds new capabilities to create versions, name versions, and compute on version data. Applications can focus versioning where and when it is most productive, and customize for each application structure independently. This control is portable, and its embedding in application source makes it natural to express and easy to maintain. The ability to name multiple versions and “partially materialize” them efficiently makes ambitious forward-recovery based on “data slices” across versions or data structures both easy to express and efficient. Using several large applications (OpenMC, preconditioned conjugate gradient (PCG) solver, ddcMD, and Chombo), we evaluate the programming effort to add resilience. The required changes are small (< 2% lines of code (LOC)), localized and machine-independent, and perhaps most important, require no software architecture changes. We also measure the overhead of adding GVR versioning and show that overheads < 2% are generally achieved. This overhead suggests that GVR can be implemented in large-scale codes and support portable error recovery with modest investment and runtime impact. Our results are drawn from both IBM BG/Q and Cray XC30 experiments, demonstrating portability. We also present two case studies of flexible error recovery, illustrating how GVR can be used for multi-version rollback recovery, and several different forward-recovery schemes. GVR’s multi-version enables applications to survive latent errors (silent data corruption) with significant detection latency, and forward recovery can make that recovery extremely efficient. Lastly, our results suggest that GVR is scalable, portable, and efficient. GVR interfaces are flexible, supporting a variety of recovery schemes, and altogether GVR embodies a gentle-slope path to tolerate growing error rates in future extreme-scale systems.

The Trilinos Project produces, maintains and distributes a large collection of reusable, parallel scientific libraries. Docker provides container technologies that support compilation, packaging, distribution and execution of software on Linux, Mac OS and Windows systems, with emerging support for Cray platforms. In this short article we describe recent efforts to explore the potential for using Docker in a variety of settings to enhance several Trilinos Project workflows. The technical foundation for this article is presented in an Honors thesis of one of the authors.

Abstract not provided.

Abstract not provided.

Abstract not provided.

Our first purpose here is to offer to a general technical and policy audience a perspective on whether the supercomputing community should focus on improving the efficiency of supercomputing systems and their use rather than on building larger and ostensibly more capable systems that are used at low efficiency. After first summarizing our content and defining some necessary terms, we give a concise answer to this question. We then set this in context by characterizing performance of current supercomputing systems on a variety of benchmark problems and actual problems drawn from workloads in the national security, industrial, and scientific context. Along the way we answer some related questions, identify some important technological trends, and offer a perspective on the significance of these trends. Our second purpose is to give a reasonably broad and transparent overview of the related issue space and thereby to better equip the reader to evaluate commentary and controversy concerning supercomputing performance. For example, questions repeatedly arise concerning the Linpack benchmark and its predictive power, so we consider this in moderate depth as an example. We also characterize benchmark and application performance for scientific and engineering use of supercomputers and offer some guidance on how to think about these. Examples here are drawn from traditional scientific computing. Other problem domains, for example, data analytics, have different performance characteristics that are better captured by different benchmark problems or applications, but the story in those domains is similar in character and leads to similar conclusions with regard to the motivating question.

Abstract not provided.

Abstract not provided.

We describe a new high-performance conjugate-gradient (HPCG) benchmark. HPCG is composed of computations and data-access patterns commonly found in scientific applications. HPCG strives for a better correlation to existing codes from the computational science domain and to be representative of their performance. HPCG is meant to help drive the computer system design and implementation in directions that will better impact future performance improvement.

Computational Science and Engineering (CSE) software can benefit substantially from an explicit focus on quality improvement. This is especially true as we face increased demands in both modeling and software complexities. At the same time, just desiring improved quality is not sufficient. We must work with the entities that provide CSE research teams with publication venues, funding, and professional recognition in order to increase incentives for improved software quality. In fact, software quality is precisely calibrated to the expectations, explicit and implicit, set by these entities. We will see broad improvements in sustainability and productivity only when publishers, funding agencies and employers raise their expectations for software quality. CSE software community leaders, those who are in a position to inform and influence these entities, have a unique opportunity to broadly and positively impact software quality by working to establish incentives that will spur creative and novel approaches to improve developer productivity and software sustainability.

Abstract not provided.