Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!news.mel.connect.com.au!munnari.OZ.AU!news.ecn.uoknor.edu!news.eng.convex.com!hermes.oc.com!news.unt.edu!cs.utexas.edu!howland.reston.ans.net!newsfeed.internetmci.com!vixen.cso.uiuc.edu!news.ais.net!eagle.ais.net!adoane
From: adoane@eagle.ais.net (Andrew J. Doane)
Newsgroups: comp.unix.bsd.bsdi.misc
Subject: Re: Argh! BSDI/News Problem... help kindly requested
Date: 31 Jan 1996 09:32:46 GMT
Organization: American Information Systems, Inc.
Lines: 74
Message-ID: <4encvu$na1@news.ais.net>
References: <30f54d00.11463334@news> <4dh75m$cv1@INET_PRIME.comshare.com>
NNTP-Posting-Host: eagle.ais.net
X-Newsreader: TIN [version 1.2 PL2]
Patrick S. Roberts (stormyb@izzy.net) wrote:
: rhof@netinc.ca (Robert Hof) wrote:
: >I think I might be in need of a little expert advice here:
: >I'm running a BSDI2.0.1 system on a dx2-66 w/ 32MB ram.
[stuff deleted]
: I am having the same problems over here. We a carring a full feed,
: but it never catches up. We have a T-1 to our provider, and the
: machiene is a Twin P/5133 with 256mb of memory, and a BT-946c PCI
: SCSI-2 Controlller, and the drives are 7200rpm 9gig SEAGATES..... I
: can't figure the problem....
Well, I've got some bad news for you guys. I've been dealing with usenet
news servers for about three years now. SVR4, Linux, NetBSD, BSDI 2.0, to
name a few.
Our current setup is this:
Pentium 120 w/64m ram
2 PCI Buslogic cards
WindJammer Raid array with 16m ram R/W Cache (we only get about 2% hit anyway)
8 2gb 7200rpm drives in a Raid0+ configuration on the raid array
1 4gb 7200rpm drive
2 3c509 Ethernet cards
1 scsi controller talks to the Raid array, the other to the boot drive and
the 4gb drive, thus doubling throughput.
The raid array stores the aritcles and history files, the 4gb is used for
logging, user statistics, etc.
1 ethernet card talks directly to a gateway for our feeds (we receive 3
feeds and peer 100% with 2 others.)
The other ethernet card talks to a server LAN just for NFS mounting the spool
to other machines.
The bad news is : This setup is no longer fast enough. The problem with news
is not your internet link (assuming 256k-512k or better), its IO.
You *MUST* use a raid array with as many spindles as possible if you are
going to try to have a full feed and/or multiple peers.
I've also done a few custom modifications to help the poor thing. One big
one is that BSDI's version of innd out of the box doesn't override the
default user limits. If your using the box soley for news (you better be)
and you run it as 'news' (you should) you need to override the default
userlimits with 'limit' or /usr/sbin/sysctl to allow innd to use more
RAM, and to allow more 'news' processes. Late at night we go well over
100 clients reading news, and I think the default process limit per user
is ~50.
I also use a script that one of our programmers wrote in perl to watch for
stale nnrpd's and kill them off. Each client eats RAM. And INND is a RAM
HOG.
Not sure if this helps at all... you're not alone. As usenet/internet grows,
especially with AOL, Compu$erv, etc., providing access with lots of $$
to spend on hardware ... we're going to be playing catch-up.
(On a side note- Intel boxes arn't the greatest for news to begin with
IMHO, but since this is a BSDI discussion....)
/ajd/
--
_/_/ _/ _/_/_/_/
_/ _/ _/ _/ Andrew J. Doane
_/_/_/_/ _/ _/_/_/_/ Vice President, Systems
_/ _/ _/ _/ American Information Systems, Inc.
_/ _/ _/ _/_/_/_/ E-mail: adoane@ais.net, http://www.ais.net