Collective I/O tuning using analytical and machine learning models

Florin Isaila, Prasanna Balaprakash, Stefan M. Wild, Dries Kimpe, Rob Latham, Rob Ross, Paul Hovland

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

24 Scopus citations

Abstract

The optimization of parallel I/O has become challenging because of the increasing storage hierarchy, performance variability of shared storage systems, and the number of factors in the hardware and software stacks that impact performance. In this paper, we perform an in-depth study of the complexity involved in I/O autotuning and performance modeling, including the architecture, software stack, and noise. We propose a novel hybrid model combining analytical models for communication and storage operations and black-box models for the performance of the individual operations. The experimental results show that the hybrid approach performs significantly better and shows a higher robustness to noise than state-of-the-art machine learning approaches, at the cost of a higher modeling complexity.

Original languageEnglish
Title of host publicationProceedings - 2015 IEEE International Conference on Cluster Computing, CLUSTER 2015
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages128-137
Number of pages10
ISBN (Electronic)9781467365987
DOIs
StatePublished - Oct 26 2015
Externally publishedYes
EventIEEE International Conference on Cluster Computing, CLUSTER 2015 - Chicago, United States
Duration: Sep 8 2015Sep 11 2015

Publication series

NameProceedings - IEEE International Conference on Cluster Computing, ICCC
Volume2015-October
ISSN (Print)1552-5244

Conference

ConferenceIEEE International Conference on Cluster Computing, CLUSTER 2015
Country/TerritoryUnited States
CityChicago
Period09/8/1509/11/15

Keywords

  • I/O performance modeling
  • Model-based tuning
  • Statistical and analytical performance models

Fingerprint

Dive into the research topics of 'Collective I/O tuning using analytical and machine learning models'. Together they form a unique fingerprint.

Cite this