Characterizing and Approximating I/O Behavior of HDF5 Applications

N. Rajesh, G. Heber, A. Kougkas, X.-H. Sun

The International Conference for High Performance Computing, Networking, Storage and Analysis (SC'20)

Research areas:HPC

Systems & Workloads

hpc storage and i/o

Methods & Models

i/o characterization

Problems & Goals

i/o analysis

[Poster][Abstract][BibTeX][Citation]

Abstract:

By characterizing the I/O behavior of applications, we try to answer the following questions: Can we decompose an application profile into a simple sequence or ”genome” sequence that describes the application? Can we compare these different ”genomes” using a simple similarity metric? Using a simple similarity metric can we, Compare the I/O characteristics of 2 or more applications using their ”genomes”? Can we use a simpler mechanism to recreate the I/O behavior of the application? What can we infer about the original application’s behavior?