Analyzing the data behavior of parallel application for extracting performance knowledge

Tirado, Felipe; Wong, Alvaro; Rexachs, Dolores; Luque, Emilio

doi:10.1109/hpcs48598.2019.9188166

Cited by 5 publications

(7 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Previous PAS2P approaches [21] optimize the analysis stage but have a drawback of dependencies between events, thus reducing its performance. The dependencies between events occur due to the communication effected by constructing a global clock that orders the events by precedence.…”

Section: Proposed Methodologymentioning

confidence: 99%

“…The PAS2P methodology allows us to generate a model of a parallel application and automatically extract its most significant phases to create a signature whose execution lets us predict the application's performance on different parallel computers Due to the large number of communication events captured in the instrumentation stage of PAS2P, trace analysis is complex and costly, even more so when the application scales to a more significant number of processes analysis stage, some to identify performance problems and others to characterize the application better, improving prediction quality. Previous work exists, such as the PAS2P parallel analyzer module [21], which defines a parallel analysis for the PAS2P tool. The main drawback of this analysis model is its complexity and the large number of MPI messages resulting from communication due to PAS2P event synchronisation.…”

Section: Pas2pmentioning

confidence: 99%

“…Thereby, having to load these data from the swap memory considerably increases the analysis execution time, or in many cases, it is not even possible to execute it due to the restrictions of the target machine. • The parallel analyzer approach [21] has been developed using message passing to take advantage of distributed memory. This module allows us to use the PAS2P toolkit on a large scale, achieving an efficient analysis, since it divides the data analysis among all the resources and it executes using the same number of resources as the application uses for its execution.…”

Section: Pas2p Overviewmentioning

confidence: 99%

“…There are two similarity algorithms, the serial similarity method and the parallel similarity method [21]. In both cases, the concept is the same, the identification of phases in parallel applications, extracting directly from the logic trace.…”

Section: Identification Patternmentioning

confidence: 99%

“…Due to the PAS2P parallel analyzer [21] model, which includes several interprocess synchronization mechanisms, represented with red arrows in Fig. 1, the performance of the PAS2P analysis is inefficient because of the high time required.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Scalable performance analysis method for SPMD applications

et al. 2022

Self Cite

View full text Add to dashboard Cite

The analysis of parallel scientific applications allows us to understand their computational and communication behavior. One way of obtaining performance information is through performance tools. One such tool is parallel application signatures for performance prediction (PAS2P), based on parallel application repeatability, focusing on performance analysis and prediction. The same resources that execute the parallel application are used to perform its analysis, creating a machine independent model of the application and identifying its common patterns. However, the analysis is costly in terms of execution time due to the high number of synchronization communications performed by PAS2P, degrading performance as the number of processes increases. To solve this problem, we propose a model that reduces data dependency between processes, reducing the number of communications performed by PAS2P in the analysis stage and taking advantage of the characteristics of single program, multiple sata applications. Our analysis proposal allows us to decrease the analysis time by 29 times when the application scales to 256 processes, while keeping error levels below 11% in the runtime prediction. It is important to mention that the analysis time is not considerably affected by increasing the number of application processes.

show abstract

Section: Proposed Methodologymentioning

confidence: 99%

Section: Pas2pmentioning

confidence: 99%

Section: Pas2p Overviewmentioning

confidence: 99%

Section: Identification Patternmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations