In the presence of accelerated fault rates, which are projected to be the norm on future exascale systems, it will become increasingly difficult for high-performance computing (HPC) applications to accomplish useful computation. Due to the fault-oblivious nature of current HPC programming paradigms and execution environments, HPC applications are insufficiently equipped to deal with errors. We believe that HPC applications should be enabled with capabilities to actively search for and correct errors in their computations. The redundant multithreading (RMT) approach offers lightweight replicated execution streams of program instructions within the context of a single application process. However, the use of complete redundancy incurs significant overhead to the application performance. In this paper we present RedThreads, an interface that provides application-level fault detection and correction based on RMT, but applies the thread-level redundancy adaptively. We describe the RedThreads syntax and semantics, and the supporting compiler infrastructure and runtime system. Our approach enables application programmers to scope the extent of redundant computation. Additionally, the runtime system permits the use of RMT to be dynamically enabled, or disabled, based on the resiliency needs of the application and the state of the system. Our experimental results demonstrate how adaptive RMT exploits programmer insight and runtime inference to dynamically navigate the trade-off space between an application’s resilience coverage and the associated performance overhead of redundant computation.
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.
Gamble, Marc G.; Van Der Wijngaart, Rob F.; Teranishi, Keita T.; Parashar, Manish P.
This document provides a specification of Fenix, a software library compatible with the Message Passing Interface (MPI) to support fault recovery without application shutdown. The library consists of two modules. The first, termed process recovery , restores an application to a consistent state after it has suffered a loss of one or more MPI processes (ranks). The second specifies functions the user can invoke to store application data in Fenix managed redundant storage, and to retrieve it from that storage after process recovery.
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.
Gamble, Marc G.; Van Der Wijngaart, Rob F.; Teranishi, Keita T.; Parashar, Manish P.
This document provides a specification of Fenix, a software library compatible with the Message Passing Interface (MPI) to support fault recovery without application shutdown. The library consists of two modules. The first, termed process recovery , restores an application to a consistent state after it has suffered a loss of one or more MPI processes (ranks). The second specifies functions the user can invoke to store application data in Fenix managed redundant storage, and to retrieve it from that storage after process recovery.
PARMA (Distributed Asynchronous Resilient Models and ApH asynchronous many-task (AMT) rmogramming models and hardware idiosyncrasies, 2) improve application programmer interface (API) plication Ico-desiga activities into meaningful requirements for characterization and definition, accelerating the development of pARMAI APT is a rranslation layer runtime systems Am' 11 between an application-facing . The application-facing user-level iting the generic language constructs of C++ and adding parallel programs. Though the implementation of the provide the front end semantics, it is nonetheless fully embedded in the C++ language and leverages a widely supported front end fiack end in C++, inher- that facilitate expressing distributed asynchronous uses C++ constructs unfamiliar to many programmers to subset of C++14 functionality (gcc >= 4.9, clang >= 3.5, icc > = 16). The rranslation layer leverages C++ to map the user's code onto the fiack encI runtime APT. The fiack end APT is a set of abstract classes and function signatures that iuntime systenr developers must implement in accordance with the specification require- ments in order to interface with application code written to the must link to a iuntime systenr that implements the abstract mentations will be external, drawing upon existing provided in the pARMAI code distribution. IDARMAI fiack end templatO front end. Executable 1DARMA applications runtime APT. It is intended that these imple- technologies. However, a reference implementation will be The front end rranslation layer, and iback end APT are detailed herein. We also include a list of application requirements driving the specification (along with a list of the applications contributing to the requirements to date), a brief history of changes between previous versions of the specification, and summary of the planned changes in up- coming versions of the specification. Appendices walk the user through a more detailed set of examples of applications written in the PARMA front encI APII and provide additional technical details for those the interested reader.
Application resilience is a key challenge that has to be addressed to realize the exascale vision. Online recovery, even when it involves all processes, can dramatically reduce the overhead of failures as compared to the more traditional approach where the job is terminated and restarted from the last checkpoint. In this paper we explore how local recovery can be used for certain classes of applications to further reduce overheads due to resilience. Specifically we develop programming support and scalable runtime mechanisms to enable online and transparent local recovery for stencil-based parallel applications on current leadership class systems. We also show how multiple independent failures can be masked to effectively reduce the impact on the total time to solution. We integrate these mechanisms with the S3D combustion simulation, and experimentally demonstrate (using the Titan Cray-XK7 system at ORNL) the ability to tolerate high failure rates (i.e., node failures every 5 seconds) with low overhead while sustaining performance, at scales up to 262144 cores.
The ability to consistently handle faults in a distributed environment requires, among a small set of basic routines, an agreement algorithm allowing surviving entities to reach a consensual decision between a bounded set of volatile resources. This paper presents an algorithm that implements an Early Returning Agreement (ERA) in pseudo-synchronous systems, which optimistically allows a process to resume its activity while guaranteeing strong progress. We prove the correctness of our ERA algorithm, and expose its logarithmic behavior, which is an extremely desirable property for any algorithm which targets future exascale platforms. We detail a practical implementation of this consensus algorithm in the context of an MPI library, and evaluate both its efficiency and scalability through a set of benchmarks and two fault tolerant scientific applications.
Major exascale computing reports indicate a number of software challenges to meet the dramatic change of system architectures in near future. While several-orders-of-magnitude increase in parallelism is the most commonly cited of those, hurdles also include performance heterogeneity of compute nodes across the system, increased imbalance between computational capacity and I/O capabilities, frequent system interrupts, and complex hardware architectures. Asynchronous task-parallel programming models show a great promise in addressing these issues, but are not yet fully understood nor developed su ciently for computational science and engineering application codes. We address these knowledge gaps through quantitative and qualitative exploration of leading candidate solutions in the context of engineering applications at Sandia. In this poster, we evaluate MiniAero code ported to three leading candidate programming models (Charm++, Legion and UINTAH) to examine the feasibility of these models that permits insertion of new programming model elements into an existing code base.
This report provides in-depth information and analysis to help create a technical road map for developing next-generation programming models and runtime systems that support Advanced Simulation and Computing (ASC) work- load requirements. The focus herein is on asynchronous many-task (AMT) model and runtime systems, which are of great interest in the context of "Oriascale7 computing, as they hold the promise to address key issues associated with future extreme-scale computer architectures. This report includes a thorough qualitative and quantitative examination of three best-of-class AIM] runtime systems – Charm-++, Legion, and Uintah, all of which are in use as part of the Centers. The studies focus on each of the runtimes' programmability, performance, and mutability. Through the experiments and analysis presented, several overarching Predictive Science Academic Alliance Program II (PSAAP-II) Asc findings emerge. From a performance perspective, AIV runtimes show tremendous potential for addressing extreme- scale challenges. Empirical studies show an AM runtime can mitigate performance heterogeneity inherent to the machine itself and that Message Passing Interface (MP1) and AM11runtimes perform comparably under balanced conditions. From a programmability and mutability perspective however, none of the runtimes in this study are currently ready for use in developing production-ready Sandia ASC applications. The report concludes by recommending a co- design path forward, wherein application, programming model, and runtime system developers work together to define requirements and solutions. Such a requirements-driven co-design approach benefits the community as a whole, with widespread community engagement mitigating risk for both application developers developers. and high-performance computing runtime systein
In this position paper, we argue for improved fault-tolerance of an MPI code by introducing lightweight virtualization into the MPI interface. In particular, we outline key-value store semantics for MPI send/recv calls, thereby creating a far more expressive programming model. The general message passing semantics and imperative style of MPI application codes would remain essentially unchanged. However, the additional expressiblity of the programming model 1) enables the underlying transport layer to handle faulttolerance more transparently to the application developer, and 2) provides an evolutionary code path towards more declarative asynchronous programming models. The core contribution of this paper is an initial implementation of the DHARMA transport layer that provides the new, required functionality to support the MPI key-value store model.
Application resilience is a key challenge that must be ad-dressed in order to realize the exascale vision. Previous work has shown that online recovery, even when done in a global manner (i.e., involving all processes), can dramatically re-duce the overhead of failures when compared to the more traditional approach of terminating the job and restarting it from the last stored checkpoint. In this paper we suggest going one step further, and explore how local recovery can be used for certain classes of applications to reduce the over-heads due to failures. Specifically we study the feasibility of local recovery for stencil-based parallel applications and we show how multiple independent failures can be masked to effectively reduce the impact on the total time to solution.
Resilience is a major challenge for large-scale systems. It is particularly important for iterative linear solvers, since they take much of the time of many scientific applications. We show that single bit flip errors in the Flexible GMRES iterative linear solver can lead to high computational overhead or even failure to converge to the right answer. Informed by these results, we design and evaluate several strategies for fault tolerance in both inner and outer solvers appropriate across a range of error rates. We implement them, extending Trilinos’ solver library with the Global View Resilience (GVR) programming model, which provides multi-stream snapshots, multi-version data structures with portable and rich error checking/recovery. Lastly, experimental results validate correct execution with low performance overhead under varied error conditions.