What is the main point of the article?

as the size of parallel systems and the number of processorsused by individual applications rises, the traditional approach of sequentially analyzing a single global trace file becomes increasingly constrained by the large number of events.

What is the purpose of the paper?

To satisfy their increasing demand for computing power, advanced numerical simulations are required to harness larger numbers of processors offered by modern capability computing systems, such as the IBM BlueGene/L system JUBL at Forschungszentrum Jülich.

What is the time lost in the late sender?

After verifying that it relates to an N-to-N operation, accomplished by examiningthe associated region identifier, the analyzer invokes the detection algorithm, which determines the latest of the corresponding enter events using an MPI_Allreduce operation.

Why are the times reported in this diagram the best?

Due to the often considerable variation in the time for file I/O (e.g., depending on overall file-system load) the times reported are the best of several measurements.

What is the time needed for the analysis of ASC benchmark SMG2000 traces?

Figure 2 charts wall-clock times for the analysis of ASC benchmark SMG2000 traces with a range of process numbers (the 8-fold doubling of process numbers necessitates a log-log scaleto show the corresponding range of times).

Open AccessBook Chapter10.1007/11846802_43

Scalable parallel trace-based performance analysis

Q: What is the purpose of the article?

In this article, the authors outline how the pattern search can be done in a more scalable way by exploiting both distributed memory and parallel processing capabilities available on modern large-scale systems and discuss first empirical results.

Markus Geimer, +3 more

- 17 Sep 2006

- Vol. 4, Iss: 2, pp 303-312

108

TL;DR: A scalable version of the expert analysis based on analyzing separate local trace files with a parallel tool which ‘replays' the target application's communication behavior is presented.

Abstract: Automatic trace analysis is an effective method for identifying complex performance phenomena in parallel applications. However, as the size of parallel systems and the number of processors used by individual applications is continuously raised, the traditional approach of analyzing a single global trace file, as done by kojak's expert trace analyzer, becomes increasingly constrained by the large number of events. In this article, we present a scalable version of the expert analysis based on analyzing separate local trace files with a parallel tool which ‘replays' the target application's communication behavior. We describe the new parallel analyzer architecture and discuss first empirical results.

AI Agents for this Paper

Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps

Most frequently asked questions

1. What are the contributions in this paper?

In this paper, the authors outline how the pattern search can be done in a more scalable way by exploiting both distributed memory and parallel processing capabilities available on modern large-scale systems and discuss first empirical results.

2. What is the advantage of the parallel analyzer?

The parallel analyzer uses a distributed memory approach, where each process reads only the local trace data that were recorded for the corresponding process of thetarget application.

3. how long did the parallel replay take?

The parallel replay for the largest set of execution traces from 16,384 SMG2000 processes, amounting to over 40,000 million events (230 GBytes of trace files), took less then 3 minutes.

4. What is the advantage of the parallel analysis approach?

Since the size of local traces can be limited by selective tracing – i.e., by recording events only for code regions and time intervals of particular interest – the authors assume that the local trace data can be completely held in the main memory of the compute nodes.

Figure 2: Wall-clock execution times for SMG2000 analysis using the new prototype at a range of scales. Linear scaling is the bold dotted line

Figure 1: Two examples of inefficient program behavior; one for point-to-point communication (Late Sender) and one for collective operations (Wait at N x N)

Citations

•Journal Article•10.1016/J.JPDC.2008.09.001

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

Michael Noeth, +4 more

- 01 Aug 2009

TL;DR: An approach is contributed that provides orders of magnitude smaller, if not near-constant size, communication traces regardless of the number of nodes while preserving structural information.

...read moreread less

144

ScalaTrace: Scalable Compression and Replay of Communication Traces for High Performance Computing

Michael Noeth, +4 more

- 16 May 2008

TL;DR: In this article, the authors introduce intra-and inter-node compression techniques of MPI events that are capable of extracting an application's communication structure and present a replay mechanism for the traces generated by their approach and discuss results of their implementation for BlueGene/L.

...read moreread less

116

•Proceedings Article•10.1145/1375527.1375537

Preserving time in large-scale communication traces

Prasun Ratn, +3 more

- 07 Jun 2008

TL;DR: The results prove time-preserving tracing without loss of communication information can scale in the number of nodes and time steps, which is a result without precedent.

...read moreread less

Journal Article•10.1016/J.PARCO.2009.02.003

A scalable tool architecture for diagnosing wait states in massively parallel applications

Markus Geimer, +3 more

- 01 Jul 2009

TL;DR: A scalable approach is presented, based on a parallel replay of the target application's communication behavior, that can efficiently identify wait states at the previously inaccessible scale of 65,536 processes and that has potential for even larger configurations.

...read moreread less

...

Expand

References

Lecture Notes in Computer Science 2382

Petrus Bollen

- 01 Jan 2002

36.7K

VAMPIR: Visualization and Analysis of MPI Resources

Wolfgang E. Nagel, +4 more

- 04 Feb 2010

TL;DR: The powerful zooming feature of VAMPIR allows to identify problems at any level of detail and provides flexible filter operations to reduce the amount of information displayed.

...read moreread less

421

Journal Article•10.1016/S1383-7621(03)00102-4

Automatic performance analysis of hybrid MPI/OpenMP applications

Felix Wolf, +1 more

- 01 Nov 2003

TL;DR: The EXPERT performance-analysis environment provides a complete tracing-based solution for automatic performance analysis of MPI, OpenMP, or hybrid applications running on parallel computers with SMP nodes.

...read moreread less

166

VAMPIR: Visualization and Analysis of MPI Resources

Interner Bericht

- 01 Jan 1996

110

•Proceedings Article•10.5555/370049.370458

From Trace Generation to Visualization: A Performance Framework for Distributed Parallel Systems

C.E. Wu, +7 more

- 01 Nov 2000

TL;DR: A trace analysis framework is described, from trace generation to visualization, that includes a unified tracing facility on IBMâ SPä systems, a self-defining interval file format, an API for framework extensions, utilities for merging and statistics generation, and a visualization tool with preview and multiple time-space diagrams.

...read moreread less

108

...

Expand

Scalable parallel trace-based performance analysis

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What are the contributions in this paper?

2. What is the advantage of the parallel analyzer?

3. how long did the parallel replay take?

4. What is the advantage of the parallel analysis approach?

5. What is the main point of the article?

6. What is the purpose of the paper?

7. What is the time lost in the late sender?

8. What is the purpose of the article?

9. Why are the times reported in this diagram the best?

10. What is the time needed for the analysis of ASC benchmark SMG2000 traces?

Figures

Citations

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

ScalaTrace: Scalable Compression and Replay of Communication Traces for High Performance Computing

Usage of the SCALASCA toolset for scalable performance analysis of large-scale parallel applications

Preserving time in large-scale communication traces

A scalable tool architecture for diagnosing wait states in massively parallel applications

References

Lecture Notes in Computer Science 2382

VAMPIR: Visualization and Analysis of MPI Resources

Automatic performance analysis of hybrid MPI/OpenMP applications

VAMPIR: Visualization and Analysis of MPI Resources

From Trace Generation to Visualization: A Performance Framework for Distributed Parallel Systems

Related Papers (5)

The Tau Parallel Performance System

Automatic performance analysis of hybrid MPI/OpenMP applications

VAMPIR: Visualization and Analysis of MPI Resources

A Portable Programming Interface for Performance Evaluation on Modern Processors

An algebra for cross-experiment performance analysis