*BSD News Article 50854


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!news.uwa.edu.au!classic.iinet.com.au!news.uoregon.edu!vixen.cso.uiuc.edu!howland.reston.ans.net!newsfeed.internetmci.com!news.mathworks.com!tank.news.pipex.net!pipex!news.sprintlink.net!news.magicnet.net!rd.gencon.com!usenet
From: rd@tarpit.rd.gencon.com (Bob Thrush)
Newsgroups: comp.unix.bsd.bsdi.misc
Subject: bsdi 2.0 news disk performance
Date: 08 Sep 1995 19:00:39 GMT
Organization: Genesis Controls, Inc.
Lines: 50
Message-ID: <RD.95Sep8150039@tarpit.rd.gencon.com>
NNTP-Posting-Host: localhost.rd.gencon.com
To: bsdi-users@bsdi.com

  Help! Can someone help me tune this news system to achieve better
copy throughput?  Or suggest a better method to copy news file
systems?

  I've had a few occasions (through crashes and setting up a slave
news server) that involve copying an active news spool to a freshly
newfs'd drive.  Each time I've noticed what I consider low disk
performance while performing the copy on either bsdi 2.0 or 2.0.1.
I'm currently building an additional server and am copying the spool
from 2 2GB drives to one 9GB drive.  Throughput seems to average ~2.5
MB/minute during the copy operation which is done by:

cd /news
pax -rw -pe spool /news2
cd /news/.0
pax -rw -pe alt /news2/spool

  Here's some additional data points:
  The kernel has been configured with "maxusers 64" and 
"options CHILD_MAX=256".  Hardware consists of a 486/66DX2 w/ 32MB
RAM and a BusLogic 946 PCI controller.

  The disks are all on a single internal cable and are arranged as
follows:

sd0 = system disk + news history
sd1 = news over.view
sd2 = /news     - all but alt
sd3 = /news/.0  - alt
sd4 = /news2

  sd0 & sd1 are Fujitsu 1GB (M1606S-512) and sd2 and sd3 are
Micropolis 2GB (4221-09SC21020AV).  sd4 will become the news spool and
is a Micropolis 9GB (1991-27MZ 1020S).  The sd4 filesystem was created
by:

newfs -i 2800 -b 4096 -f 1024 -a 8 -d 0 -m 5 /dev/rsd4h

which resulted in the following (superblock backup list deleted):

Warning: 368 sector(s) in last cylinder unallocated
/dev/rsd4h:     17550992 sectors in 8570 cylinders of 1 tracks, 2048 sectors
        8569.8MB in 536 cyl groups (16 c/g, 16.00MB/g, 5632 i/g)

  Is there something I've overlooked that could have been done to
improve on the 2.5 MB / minute copy performance?

Thanks, Bob
-- 
Bob Thrush rd@rd.gencon.com    407 425-4845/ FAX 841-8004