I/O performance evaluation of parallel scientific applications in a Cloud Environment

TitleI/O performance evaluation of parallel scientific applications in a Cloud Environment
Publication TypeConference Proceedings
Year of Conference2015
AuthorsGomez-Sanchez, P, Méndez, S, Rexachs, D, Luque, E
EditorArabnia, HR, Joe, K, Shouno, H, Ishii, H, Nishikawa, H
Conference NameInternational Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA)
Date Published07/2015
PublisherWORLDCOMP’15. CSREA Press
Conference LocationAthens, Greece.
ISBN Number1-60132-400-6

Performance evaluation of parallel application plays an important role on High Performance Computing (HPC), in cloud computing as well as in the classical systems. As the scientific community selects the cloud environment to access resources of HPC, performance evaluation brings new challenges such as defining methods to analyze the performance on a more configurable environment than traditional cluster computers. In this paper, we present a methodology to evaluate the I/O performance of parallel application over the Virtual Clusters in a cloud environment. We represent the I/O requirements of the parallel application in an I/O model and define a percentage required of the memory and storage to determine if the application is I/O intensive. If the application is I/O intensive, we define boundaries for the data transfer rate to identify if the application is limited by the I/O subsystem. We have evaluated on different Virtual Clusters configurations for the kernels of scientific applications S3DIO and BT-IO. Experimental results show that our methodology allows us to identify when an application is I/O intensive and to evaluate if it is limited by the I/O subsystem.

Campus d'excel·lència internacional U A B