System Issues in Implementing High Speed
Distributed Parallel Storage Systems
Brian Tierney (bltierney@lbl.gov)
Bill Johnston (wejohnston@lbl.gov)
Hanan Herzog, Gary Hoo, Guojun Jin, Jason Lee
Imaging Technologies Group
Lawrence Berkeley Laboratory
Berkeley, CA 94720
Abstract
In this paper we describe several aspects of implementing a high speed
network-based distributed application. We describe the design and
implementation of a distributed parallel storage system that uses high
speed ATM networks as a key element of the architecture. The
architecture provides what amounts to a collection of network-based
disk block servers, and an associated name server that provides some
file system functionality. The implementation approach is that of user
level software that runs on UNIX workstations. Both the architecture
and the implementation are intended to provide for easy and economical
scalability in both performance and capacity. We describe the software
architecture, the implementation and operating system overhead issues,
and our experiences with this approach in an IP-over-ATM WAN. Although
most of the paper is specific to a distributed parallel data server,
we believe many of the issues we encountered are generally applicable
to any high speed network-based application.
Download the full text of this paper in
ASCII (37,336 bytes) and
POSTSCRIPT (142,570 bytes) form.
To Become a USENIX Member, please see our
Membership Information.