Portable profiling and tracing for parallel, scientific applications using C++

Shende, Sameer; Malony, Allen D.; Cuny, Janice E.; Beckman, Peter H.; Karmesin, Steve; Lindlan, Kathleen

doi:10.1145/281035.281049

Cited by 72 publications

(35 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Hence, structural information and event ordering are not preserved. There are many other tools [1,26,28,12] that report aggregate information, often based on the profiling layer of MPI, as is the case with mpiP or produce terabytes of traces. None of these tools are suitable for lossless tracing and later replay in a scalable manner with a single trace file in the megabyte range.…”

Section: Related Workmentioning

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

113

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: Related Workmentioning

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

113

111

View full text Add to dashboard Cite

show abstract

“…The TAU performance system (Shende et al 1998;Malony and Shende 2000; University of Oregon b) is designed as a tool framework, whereby tool components and modules are integrated and coordinate their operation using well- defined interfaces and data formats. The TAU framework architecture is organized into three layers -instrumentation, measurement, and analysis -where within each layer multiple modules are available and can be configured in a flexible manner under user control.…”

Section: Tau Performance System Architecturementioning

confidence: 99%

“…Phases can also be nested, in which case profiling follows normal scoping rules and is associated with the closest parent phase obtained by traversing up the callstack. Phases should not overlap, as it also true for interval events (Shende et al 1998). Each thread of execution in an application has a default phase and this corresponds to the top level event.…”

Section: Phase Profilingmentioning

confidence: 99%

The Tau Parallel Performance System

Shende

Malony

2006

The International Journal of High Performance Computing Applica

Self Cite

924

445

View full text Add to dashboard Cite

The ability of performance technology to keep pace with the growing complexity of parallel and distributed systems depends on robust performance frameworks that can at once provide system-specific performance capabilities and support high-level performance problem solving. Flexibility and portability in empirical methods and processes are influenced primarily by the strategies available for instrumentation and measurement, and how effectively they are integrated and composed. This paper presents the TAU (Tuning and Analysis Utilities) parallel performance system and describe how it addresses diverse requirements for performance observation and analysis.

show abstract

“…Profiling [7,8] captures the behavior of an application by monitoring its execution. Monitoring can be based on hardware counter sampling or it can require the instrumentation of the application's source code or its binary executable.…”

Section: Background and Motivationmentioning

confidence: 99%

The Tracefile Testbed: a community repository for identifying and retrieving HPC performance data

Ferschweiler¹,

Harrah²,

Keon³

et al. 2005

IJHPCN

View full text Add to dashboard Cite

Background and motivationA high-performance computing (HPC) application is characterized by many variables that control its execution and determine its performance. Variables such as algorithm type, problem size, input parameters, programming languages and paradigms, libraries, hardware architecture, etc., can have very significant effects on program behavior. It is important to understand the role played by each variable and the ways they combine to influence the performance achieved, or achievable, by the application.Two approaches are commonly used for the purpose of understanding these effects: performance profiling and performance prediction. Profiling [7,8] captures the behavior of an application by monitoring its execution. Monitoring can be based on hardware counter sampling or it can require the instrumentation of the application's source code or its binary executable. The data produced by monitoring may be analyzed on-the-fly or stored as tracefiles for post-mortem analysis. Many of the tools currently available for HPC performance analysis are based on tracefiles. Examples include:• These techniques attempt to provide estimates of the performance achievable by an application by analyzing its structure and the influences of compiler transformations and the system architecture, using symbolic analysis, simulation, or other model-based methods. Prediction tools often rely directly or indirectly on tracefiles. The data from tracefiles can serve as the basis for constructing or validating the performance model, or can be used directly by the tool to adjust the model to the characteristics of a particular application (e.g., [9]).Tracefiles are typically generated by the application programmer as part of the performance tuning process. Our field studies of HPC programmers indicate that many experienced programmers also create suites of simple pseudo-benchmark codes and generate tracefiles to help establish basic performance characteristics when they

show abstract

Portable profiling and tracing for parallel, scientific applications using C++

Cited by 72 publications

References 15 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

The Tau Parallel Performance System

The Tracefile Testbed: a community repository for identifying and retrieving HPC performance data

Contact Info

Product

Resources

About