Characterization of MPI Usage on a Production Supercomputer

Chunduri, Sudheer; Parker, Scott; Balaji, Pavan; Harms, Kevin; Kumaran, Kalyan

doi:10.1109/sc.2018.00033

Cited by 62 publications

(39 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To characterize the performance of MPI collectives, we set the threshold value of MPI runtime parameter to be within 1 B (the minimum) and 1 MB (the maximum). In NPB and HPCC, each benchmark contains various MPI collectives, and the ratio of the collective running time to the total execution time varies [8], which means that the effection of collectives to the general performance is different.…”

Section: Profiling Setupmentioning

confidence: 99%

“…But when the number of processes is more than 64, our approach can significantly improve the performance. When the message size is 8 KB and the number of processes is 1024, the parameters are configured to be (8,13), and the speedup reaches its largest, which is around 31.7%.…”

Section: Model Evaluationmentioning

confidence: 99%

“…the number of processes is 1024, the parameters are configured to be (8,12), with an aim to achieve the largest performance improvement (i.e., 40.7%).…”

Section: Model Evaluationmentioning

confidence: 99%

“…Based on the knowledge extracted from the microbenchmarks, we then develop a model to automatically choose, from a set of available algorithms available to an operation, which communication algorithm to use at runtime. Our key insight is that most real-world applications contain various MPI collective operations [8], and each of their bestperforming communication algorithms is largely depending on the message size. If we can find, from microbenchmarks, the correlation between the message size and the optimal operation-specific algorithm, we can then apply and transfer this knowledge to optimize new, unseen MPI programs.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Auto-Tuning MPI Collective Operations on Large-Scale Parallel Systems

Zheng

Fang

Chen

et al. 2019

2019 IEEE 21st International Conference on High Performance Computing and Communications; IEEE 17th International Conference On

View full text Add to dashboard Cite

MPI libraries are widely used in applications of high performance computing. Yet, effective tuning of MPI colletives on large parallel systems is an outstanding challenge. This process often follows a trial-and-error approach and requires expert insights into the subtle interactions between software and the underlying hardware. This paper presents an empirical approach to choose and switch MPI communication algorithms at runtime to optimize the application performance. We achieve this by first modeling offline, through microbenchmarks, to find how the runtime parameters with different message sizes affect the choice of MPI communication algorithms. We then apply the knowledge to automatically optimize new unseen MPI programs. We evaluate our approach by applying it to NPB and HPCC benchmarks on a 384-node computer cluster of the Tianhe-2 supercomputer. Experimental results show that our approach achieves, on average, 22.7% (up to 40.7%) improvement over the default setting.

show abstract

Section: Profiling Setupmentioning

confidence: 99%

Section: Model Evaluationmentioning

confidence: 99%

“…the number of processes is 1024, the parameters are configured to be (8,12), with an aim to achieve the largest performance improvement (i.e., 40.7%).…”

Section: Model Evaluationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Auto-Tuning MPI Collective Operations on Large-Scale Parallel Systems

Zheng

Fang

Chen

et al. 2019

2019 IEEE 21st International Conference on High Performance Computing and Communications; IEEE 17th International Conference On

View full text Add to dashboard Cite

show abstract

“…For example, an operating system can interrupt processors at any time for its activities, causing detours in computations and degrading performance (Ferreira et al, 2008; Hoefler et al, 2010). Inter-job contention for shared resources such as network bandwidth, routers, and links is another source of variability that can affect application performance (Chunduri et al, 2017; Parker et al, 2017). Developing coherent performance models in the presence of variability is difficult (Beckman et al, 2006; Hoefler et al, 2007), particularly between systems.…”

Section: Introductionmentioning

confidence: 99%

Understanding performance variability in standard and pipelined parallel Krylov solvers

Morgan

Sanan

Knepley

et al. 2020

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

In this work, we collect data from runs of Krylov subspace methods and pipelined Krylov algorithms in an effort to understand and model the impact of machine noise and other sources of variability on performance. We find large variability of Krylov iterations between compute nodes for standard methods that is reduced in pipelined algorithms, directly supporting conjecture, as well as large variation between statistical distributions of runtimes across iterations. Based on these results, we improve upon a previously introduced nondeterministic performance model by allowing iterations to fluctuate over time. We present our data from runs of various Krylov algorithms across multiple platforms as well as our updated non-stationary model that provides good agreement with observations. We also suggest how it can be used as a predictive tool.

show abstract

Understanding the use of message passing interface in exascale proxy applications

Sultana

Rüfenacht

Skjellum

et al. 2020

Concurrency and Computation

View full text Add to dashboard Cite

Summary The Exascale Computing Project (ECP) focuses on the development of future exascale‐capable applications. Most ECP applications use the message passing interface (MPI) as their parallel programming model with mini‐apps serving as proxies. This paper explores the explicit usage of MPI in such ECP proxy applications. We empirically analyze 14 proxy applications from the ECP Proxy Apps Suite. We use the MPI profiling interface (PMPI) to collect MPI usage patterns in ECP proxy apps. Our analysis shows that a small subset of features from MPI is commonly used in the proxies of exascale‐capable applications, even when they reference third‐party libraries. This study is intended to provide a better understanding of the use of MPI in current exascale applications. The findings can help focus software investments made for exascale systems in the MPI middleware including optimization, fault‐tolerance, tuning, and hardware‐offload.

show abstract

Characterization of MPI Usage on a Production Supercomputer

Cited by 62 publications

References 18 publications

Auto-Tuning MPI Collective Operations on Large-Scale Parallel Systems

Auto-Tuning MPI Collective Operations on Large-Scale Parallel Systems

Understanding performance variability in standard and pipelined parallel Krylov solvers

Understanding the use of message passing interface in exascale proxy applications

Contact Info

Product

Resources

About