Argonne National Laboratory

Improving Parallel I/O Autotuning with Performance Modeling

TitleImproving Parallel I/O Autotuning with Performance Modeling
Publication TypeConference Paper
Year of Publication2014
AuthorsBehzad, B, Byna, S, Wild, SM, Prabhat, A, Snir, M
Conference NameHigh-Performance Parallel and Distributed Computing (HPDC14)
PublisherAssociation for Computing Machinery
Conference LocationVancouver, Canada
Other NumbersANL/MCS-P5066-0114

Various layers of the parallel I/O subsystem offer tunable parameters for improving I/O performance on large-scale computers. However, searching through a large parameter space is challenging. We are working towards an autotun- ing framework for determining the parallel I/O parameters that can achieve good I/O performance for different data write patterns. In this paper, we characterize parallel I/O and discuss the development of predictive models for use in effectively reducing the parameter space. We have ap- plied our models to two different I/O kernels, VPIC-IO and VORPAL-IO, that are derived from real, large-scale simu- lation codes. We have tested the framework on the Cray XE6 Hopper system and our results show that the search time can be reduced from 12 to 2 hours, while still achieving upto 54X I/O performance advantage over system default settings.