Florin Isaila
florin@ira.uka.de
Walter F. Tichy
University of Karlsruhe
This paper presents Clusterfile, a file system that
provides parallel file access on a cluster of computers.
Existing parallel file systems offer little control
over matching I/O access patterns and file data layouts.
Without a suitable matching, however, applications may
face the following problems: contention at I/O nodes,
fragmentation of file data, false sharing, numerous,
small network messages, and high overhead for scat-
tering/gathering the data. Clusterfile addresses some
of these issues. With Clusterfile, parallel applications
can fully control the layout of a file over a cluster and
can tailor it to access patterns. They can also set
arbitrary views on a file. Views hide the parallel
structure of the file and ease the programmer's burden
of computing complex access indices. The mapping between
views and layouts are pre-computed automatically,
optimizing disk accesses. Clusterfile uses the same
internal representation for specifying file layouts,
access patterns, and the mappings between them.