From owner-freebsd-hackers  Wed Jun 17 20:08:01 1998
Return-Path: <owner-freebsd-hackers@FreeBSD.ORG>
Received: (from majordom@localhost)
          by hub.freebsd.org (8.8.8/8.8.8) id UAA02904
          for freebsd-hackers-outgoing; Wed, 17 Jun 1998 20:08:01 -0700 (PDT)
          (envelope-from owner-freebsd-hackers@FreeBSD.ORG)
Received: from bandicoot.prth.tensor.pgs.com (bandicoot.prth.tensor.pgs.com [157.147.224.1])
          by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id UAA02876
          for <hackers@freebsd.org>; Wed, 17 Jun 1998 20:07:53 -0700 (PDT)
          (envelope-from shocking@ariadne.prth.tensor.pgs.com)
Received: from ariadne.tensor.pgs.com (ariadne [157.147.227.36])
	by bandicoot.prth.tensor.pgs.com (8.8.8/8.8.8) with SMTP id LAA07321
	for <hackers@freebsd.org>; Thu, 18 Jun 1998 11:07:06 +0800 (WST)
Received: from ariadne by ariadne.tensor.pgs.com (SMI-8.6/SMI-SVR4)
	id LAA27243; Thu, 18 Jun 1998 11:07:21 +0800
Message-Id: <199806180307.LAA27243@ariadne.tensor.pgs.com>
X-Mailer: exmh version 2.0.2 2/24/98
To: hackers@FreeBSD.ORG
Subject: Clusters, Distributed File Systems and the like.
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Thu, 18 Jun 1998 11:07:21 +0800
From: Stephen Hocking-Senior Programmer PGS Tensor Perth <shocking@ariadne.prth.tensor.pgs.com>
Sender: owner-freebsd-hackers@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG


I'm wondering if anyone has done some work on a distributed file system, where 
various nodes are each responsible for part of a filesystem and can see other 
node's part of the filesystem, with files being able to span more than one 
node. I think Greg Lehey's work (vinum) could be used as part of a solution.

The reason I ask is that I'm working for a geophysics data processing company, 
and we have all the 32 & 64 node IBM SP2 boxes floating about. Each node runs 
its own copy of AIX and has a couple of 9Gb disks, of which one is for the OS 
and the other is for the CFS (Common File System) which is shared between all 
the nodes in a system. It struck me as an interesting problem which has 
probably been solved a number of times. I'm in the process of porting our 
geophysical software to FreeBSD, which currently uses PVM and will soon be 
using MPI to distribute the processes among the nodes. I'm not expecting the 
performance to be within cooee of the SP2s, but it's an interesting exercide 
all the same.


	Stephen


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message