Hardware Reference
In-Depth Information
10.1 Motivation
OrangeFS is the next evolution of the Parallel Virtual File System (PVFS),
an open-source high performance cluster file system. OrangeFS delivers high
performance parallel access for a wide array of applications ranging from en-
gineering/scientic to advanced computing and emerging \big data" applica-
tions. The goals of OrangeFS are:
Run at user level
Allow parallel access to data and metadata
Minimize bottlenecks that limit parallelism
Provide a diverse set of client interfaces
OrangeFS can be run entirely at user level, or a small kernel module can be
used to mount it like any other file system under Linux, Windows, and OSX
for the use of standard programs and utilities. Data and metadata are trans-
parently distributed across multiple servers and can be accessed in parallel by
multiple clients. OrangeFS relaxes POSIX consistency semantics where neces-
sary to eliminate bottlenecks while offering a global POSIX namespace. Client
interfaces include mounting via the OS kernel, MPI-IO, Hadoop JNI Client,
and WebDav support.
A large community contributes to ongoing OrangeFS development, includ-
ing a staff of professional developers who support OrangeFS, improve stability
and functionality for the base system, and develop new features and interfaces.
Professional support, development, and documentation make OrangeFS a su-
perior high performing open-source file system for a growing range of scientific,
advanced computing, and \big data" applications.
10.1.1 PVFS1
PVFS1 was developed in 1993 for Parallel Virtual Machine (PVM) as part
of a NASA grant. In 1994, the software was rewritten to use TCP/IP, targeted
for a cluster of Digital Equipment Corp. Alpha workstations networked using
switched FDDI, PVFS1 striped data across multiple servers and allowed I/O
requests based on a file view that described a strided access pattern. Striping
and file view were independent of common record size.
PVFS1 [4] was ported to Linux and featured on the first Beowulf computer
at Goddard Space Flight Center [3]. In 1997 at a cluster meeting in Pasadena,
California, the developing group agreed that PVFS should be released as an
open-source package [5].
