A Framework for Performance Modeling and Prediction

Snavely, Allan; Carrington, Laura; Wolter, Nicole; Labarta, Jesús; Badía, Rosa M.; Purkayastha, Avi

doi:10.1109/sc.2002.10004

Cited by 153 publications

(145 citation statements)

References 7 publications

Supporting

Mentioning

141

Contrasting

Order By: Relevance

“…First are low-level "probes" that measure the rates at which a machine can perform fundamental operations. Examples of this class include MAPS [10], [11], STREAM [12], the Intel MPI Benchmark [13] (formerly Pallas PMB) and SKAMPI [14] MPI benchmarks, and to some extent the LINPACK benchmark [15]. In the specific context of probes for Grid platforms, we find the GrASP project [2], which we describe and use in this work, and other projects such as RGRBench [16] that measures the performance of Grid information service and was used in [17].…”

Section: Related Workmentioning

confidence: 99%

Measuring the Performance and Reliability of Production Computational Grids

Khalili

Olschanowsky

et al. 2006

2006 7th IEEE/ACM International Conference on Grid Computing

View full text Add to dashboard Cite

Abstract-In this work we report on data gathered via a deployment of a monitoring and benchmarking infrastructure on two production grid platforms, TeraGrid and Geon. Our result show that these production grids are rather unavailable, with success rates for benchmark and application runs between 55% and 80%. We also found that performance fluctuation was in the 50% range, expectedly mostly due to batch schedulers.We also investigate whether the execution time of a typical grid application can be predicated based on previous runs of simple benchmarks. Perhaps surprisingly, we find that application execution time can be predicted with a relative error as low as 9%.

show abstract

Section: Related Workmentioning

confidence: 99%

Measuring the Performance and Reliability of Production Computational Grids

Khalili

Olschanowsky

et al. 2006

2006 7th IEEE/ACM International Conference on Grid Computing

View full text Add to dashboard Cite

show abstract

“…When the performance skeleton of an application is available, an estimate of the application execution time in a new environment is obtained by simply executing the performance skeleton and appropriately scaling the measured The state of the art in performance prediction and scheduling for distributed computing environments is based on modeling of application characteristics and execution environments, with some example systems discussed in [4,5,6]. The research presented in this paper is fundamentally different in being based on synthetically generated executable code as the primary vehicle for performance prediction.…”

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

“…In such scenarios, communication traces often provide the insight to detect inefficiencies and help in problem tuning [21], [4]. Traces are also utilized to drive HPC simulations to determine the effect of interconnect changes for future procurements [17], [25], [14], [19], [26].…”

Section: Introductionmentioning

confidence: 99%

ACURDION: An adaptive clustering-based algorithm for tracing large-scale MPI applications

Bahmani¹,

Mueller²

2015

2015 IEEE International Conference on Big Data (Big Data)

View full text Add to dashboard Cite

Abstract-Communication traces help developers of highperformance computing (HPC) applications understand and improve their codes. When run on large-scale HPC facilities, the scalability of tracing tools becomes a challenge. To address this problem, traces can be clustered into groups of processes that exhibit similar behavior. Instead of collecting traces information of each individual node, it then suffices to collect a trace of a small set of representative nodes, namely one per cluster. However, clustering algorithms themselves need to have low overhead, be scalable, and adapt to application characteristics. We devised an adaptive clustering algorithm for large-scale applications called ACURDION that traces the MPI communication of code with O(log P) time complexity where P is the number of processes. First, ACURDION identifies the parameters that differ across processes by using a logarithmic algorithm called Adaptive Signature Building. Second, it clusters the processes based on those parameters. Experiments show that collecting traces of just nine nodes/clusters suffices to capture the communication behavior of all nodes while retaining sufficient accuracy of trace events and parameters. In summary, ACURDION improves trace scalability and automation over prior approaches.

show abstract

A Framework for Performance Modeling and Prediction

Cited by 153 publications

References 7 publications

Measuring the Performance and Reliability of Production Computational Grids

Measuring the Performance and Reliability of Production Computational Grids

Construction and Evaluation of Coordinated Performance Skeletons

ACURDION: An adaptive clustering-based algorithm for tracing large-scale MPI applications

Contact Info

Product

Resources

About