| Abstract |
The compute nodes of modern high performance computing clusters are
composed of some of the fastest computing hardware available. The world's
fastest clusters are capable of 100s of teraflops, however,
with insufficient access to data or insufficient bandwidth for the
storage of results, all of that massive computing power could sit idling
in long wait states while executing I/O bound processes. In this seminar
I will survey the designs, strengths, weaknesses and capabilities of four
contemporary clustered parallel file systems designed for use with high
performance computing clusters: the Panasas Storage Cluster and its
ActiveScale file system, the Lustre file system, the Parallel Virtual
File System 2 (PVFS2) and the Red Hat Global File System (GFS).
|
