*BSD News Article 3100


Return to BSD News archive

Path: sserve!manuel!munnari.oz.au!metro!usage!sphinx!pta!yarra!melba.bby.oz.au!gnb
From: gnb@duke.bby.com.au (Gregory N. Bond)
Newsgroups: comp.unix.bsd
Subject: cpio/tar over NFS (was Re: Crashing the System)
Message-ID: <1992Aug5.023113.2506@melba.bby.oz.au>
Date: 5 Aug 92 02:31:13 GMT
References: <sxjcb-310792134054@sxjcb.uacn.alaska.edu>
Sender: usenet@melba.bby.oz.au (news READER id)
Followup-To: comp.unix.bsd
Organization: Burdett, Buckeridge & Young, Melbourne, Australia
Lines: 28
In-Reply-To: sxjcb@orca.alaska.edu's message of 31 Jul 92 21:48:35 GMT
Nntp-Posting-Host: duke

This isn't really 386BSD-specific, but....

>>>>> On 31 Jul 92 21:48:35 GMT, sxjcb@orca.alaska.edu (Jay C. Beavers) said:
Jay> The first time I was doing two cpio -it's simultaneously with the input
Jay> coming from one NFS server and the output going to another NFS server on
Jay> the src01 dist and the etc01 dist.

NFS is very slow, especially for writes, in comparison to local disks.
In situations like this, you will (almost) always find it much faster
to do something like
	rsh -n srchost "cpio -o..... | rsh dsthost cpio -i...."  
or the tar equivalent.  This means that in addition to doing the reads
ond writes via physical disks instead of NFS (probably lowering the
load on both remote hosts), each file crosses the ethernet just once
(src->dst) rather than twice (src->pc, pc->dst).

Of course, there may be political circumstances where you can
nfs-mount disks but not run rsh (strange but true!), and this doesn't
serve as a convenient stress-test for the PC...

We now return you to your regular 'I Can't Boot' / 'I Hate USL'
threads.

Greg.
--
Gregory Bond <gnb@bby.com.au> Burdett Buckeridge & Young Ltd Melbourne Australia
``USL has never sold long distance.  You're going after the wrong men in black 
  hats.  (Or, in the case of Plan 9, black space suits)'' - Tom Limoncelli