home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.unix.ultrix
- Path: sparky!uunet!decwrl!pa.dec.com!nntpd2.cxo.dec.com!nabeth!alan
- From: alan@nabeth.enet.dec.com (Alan Rollow - Alan's Home for Wayward Tumbleweeds.)
- Subject: Re: kernel configuration recommendations?
- Message-ID: <1992Sep14.204930.13982@nntpd2.cxo.dec.com>
- Lines: 54
- Sender: alan@nabeth (Alan Rollow - Alan's Home for Wayward Tumbleweeds.)
- Reply-To: alan@nabeth.enet.dec.com (Alan Rollow - Alan's Home for Wayward Tumbleweeds.)
- Organization: Digital Equipment Corporation
- References: <1992Sep11.185331.1@qucdnee.ee.queensu.ca>
- Date: Mon, 14 Sep 1992 20:49:30 GMT
-
-
- In article <1992Sep11.185331.1@qucdnee.ee.queensu.ca>, rick@qucdnee.ee.queensu.ca writes:
- >
- >i have a lab of two dozen decstation 5000/20s (12 meg, rz23l) with a
- >5000/240 (32 meg, 2 rz57s, ultrix 4.2a) as a server. all the clients
- >page locally. are there any sensible changes to make to the server
- >kernel? (increasing bufcache comes immediately to mind - what are
- >sensible values to start with?) if there's any general advice to be had
- >i'd appreciate it a lot.
-
- The two recommendations you have so far good ones:
-
- 1. (Yours) Increase the size of the buffer cache.
-
- 2. Set cache_bufcache.
-
- For sizing the buffer cache, the best recommendation I can make is
- to use whatever is left over. What you'd want to do is monitor the
- amount of memory used over some typical period of usage. Based on
- this you can choose a size that will use most of what's available,
- but leave enough for typical short falls. Non-typical short falls
- will still cause the server to page, but if they happen very often
- they aren't "non-typical" and you re-adjust the size of the cache.
-
- Monitor is good for this sort of performance monitoring and has
- an example report filter included that looks at the memory usage.
- The compress Monitor tar(5) archive is on:
-
- gatekeeper.dec.com:/pub/DEC/monitor.tar.Z
-
- The other thing to watch is disk I/O. If one disk is getting all
- of the work, you might want to think about rearrange the file systems
- to spread the I/O load out over the two. If you don't have Prestoserve
- on the server, it will help a lot to improve write performance from the
- clients back to the server.
-
- The last part is the hardest, but if you can identify the typical
- file sizes being transfered and find that they are large, the I/O
- enhancement support in V4.3 may be helpful, once you have the server
- file systems setup reasonably.
-
- >
- >thanks
-
- You're welcome.
-
- >
- >rp
- >=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
- >Rick Pim bitnet:rick@qucdnee
- >
- --
- Alan Rollow alan@nabeth.cxo.dec.com
-
-