Measurement-Based Characterization of Parallel I/O Systems

This research addresses the issue of evaluating the performance of parallel I/O subsystems for a variety of I/O intensive workloads on massively parallel systems. First, a model-based workload specification methodology for generating parallel I/O intensive workloads is described. Second, a synthetic workload generator and a performance measurement and monitoring environment based on the synthetic workload generator is described. Finally, the performances of three different kinds of I/O configurations, based on a client/server model, are characterized on two production class parallel file systems. In light of the measurements obtained on two parallel file systems, possible extensions are suggested to improve the overall architecture as well as the parallel file system of the massively parallel systems.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• Sharma, Sanjay

Contributors dc:contributor

• Iyer, Ravishankar K.

• Computer Science

Language dc:language

eng

Identifier

(MiAaPQ)AAI9921729

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/81940