Poster reception---Scalable compression and replay of communication traces in massively parallel environments

Noeth, Michael; Marathe, Jaydeep; Mueller, Frank; Schulz, Martin; Supinski, Bronis de

doi:10.1145/1188455.1188605

Cited by 35 publications

(55 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our previous work, we observed that for our inter-node compression techniques the applications mapped into these three categories as follows [20]: DT, EP, and IS exhibited near-constant trace sizes irrespective of number of tasks; LU and MG showed sub-linear scaling with number of tasks; and BT, CG, and FT resulted in faster growing non-scalable traces sizes. Using our second generation algorithm describe in Sections 2 and 3) we significantly improved these results further, with the exception of IS, as described below.…”

Section: Trace Sizes and Memory Requirementsmentioning

confidence: 93%

See 1 more Smart Citation

ScalaTrace: Scalable compression and replay of communication traces for high-performance computing

Noeth

Ratn

Mueller

et al. 2009

Journal of Parallel and Distributed Computing

Self Cite

117

111

View full text Add to dashboard Cite

Characterizing the communication behavior of large-scale applications is a difficult and costly task due to code/system complexity and long execution times. While many tools to study this behavior have been developed, these approaches either aggregate information in a lossy way through high-level statistics or produce huge trace files that are hard to handle.We contribute an approach that provides orders of magnitude smaller, if not near-constant size, communication traces regardless of the number of nodes while preserving structural information. We introduce intra-and inter-node compression techniques of MPI events that are capable of extracting an application's communication structure. We further present a replay mechanism for the traces generated by our approach and discuss results of our implementation for BlueGene/L. Given this novel capability, we discuss its impact on communication tuning and beyond. To the best of our knowledge, such a concise representation of MPI traces in a scalable manner combined with deterministic MPI call replay are without any precedent.Key words: High-Performance Computing, Scalability, Communication Tracing PACS: 07.05.Bx An earlier version of this paper appeared at IPDPS'07 [20]. This journal version extends the earlier paper by novel domain-specific intra-and inter-node compression techniques, a completely redesigned inter-node merge algorithm, novel results with a larger class of codes resulting in near-constant trace sizes, a study to identify the timestep loop and extended related work.

show abstract

Section: Trace Sizes and Memory Requirementsmentioning

confidence: 93%

“…A first-generation algorithm described in prior work [20] traversed both master and slave queues. If a subsequence of events between the queues matched, the slave events were combined with the master events by merging participant lists (node IDs of master and slave).…”

Section: Location-independent Encodingsmentioning

confidence: 99%

ScalaTrace: Scalable compression and replay of communication traces for high-performance computing

Noeth

Ratn

Mueller

et al. 2009

Journal of Parallel and Distributed Computing

Self Cite

117

111

View full text Add to dashboard Cite

show abstract

“…However, simple algorithms to achieve this are at least quadratic in trace length and hence impractical for long traces. A practical tradeoff is to limit the window size for substring matching, which again risks missing long span loops [9].…”

Section: Trace Compressionmentioning

confidence: 99%

“…The underlying problems addressed in skeleton construction have many aspects in common with on-the-fly trace compression methods, in particular the work presented in [9,10]. However, the approaches are algorithmically different; specifically the approach presented performs logicalization first and compresses only the logical trace.…”

Section: Introductionmentioning

confidence: 99%

Construction and Evaluation of Coordinated Performance Skeletons

Xu¹,

Subhlok²

2008

High Performance Computing - HiPC 2008

View full text Add to dashboard Cite

Abstract. Performance prediction is particularly challenging for dynamic environments that cannot be modeled well due to reasons such as resource sharing and foreign system components. The approach to performance prediction taken in this work is based on the concept of a performance skeleton which is a short running program whose execution time in any scenario reflects the estimated execution time of the application it represents. The fundamental technical challenge addressed in this paper is the automatic construction of performance skeletons for parallel MPI programs. The steps in the skeleton construction procedure are 1) generation of process execution traces and conversion to a single coordinated logical program trace, 2) compression of the logical program trace, and 3) conversion to an executable parallel skeleton program. Results are presented to validate the construction methodology and prediction power of performance skeletons. The execution scenarios analyzed involve network sharing, different architectures and different MPI libraries. The emphasis is on identifying the strength and limitations of this approach to performance prediction.

show abstract

“…These construct correspond to the code in Figure 1. The algorithmic details of MPI event compressions over PRSDs can be found elsewhere [7,8].…”

Section: Lossless Tracingmentioning

confidence: 99%

ScalaTrace: Tracing, Analysis and Modeling of HPC Codes at Scale

Mueller

Schulz

et al. 2012

Applied Parallel and Scientific Computing

Self Cite

View full text Add to dashboard Cite

Characterizing the communication behavior of largescale applications is a difficult and costly task due to code/system complexity and their long execution times. An alternative to running actual codes is to gather their communication traces and then replay them, which facilitates application tuning and future procurements. While past approaches lacked lossless scalable trace collection, we contribute an approach that provides orders of magnitude smaller, if not near constant-size, communication traces regardless of the number of nodes while preserving structural information. We introduce intra-and inter-node compression techniques of MPI events, we develop a scheme to preserve time and causality of communication events, and we present results of our implementation for BlueGene/L. Given this novel capability, we discuss its impact on communication tuning and on trace extrapolation. To the best of our knowledge, such a concise representation of MPI traces in a scalable manner combined with time-preserving deterministic MPI call replay are without any precedence.

show abstract

Poster reception---Scalable compression and replay of communication traces in massively parallel environments

Cited by 35 publications

References 18 publications

ScalaTrace: Scalable compression and replay of communication traces for high-performance computing

ScalaTrace: Scalable compression and replay of communication traces for high-performance computing

Construction and Evaluation of Coordinated Performance Skeletons

ScalaTrace: Tracing, Analysis and Modeling of HPC Codes at Scale

Contact Info

Product

Resources

About