home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!Germany.EU.net!unido!news.Hamburg.Germany.EU.net!easix!umunk!udo
- From: udo@umunk.GUN.de (Udo Munk)
- Newsgroups: comp.os.coherent
- Subject: Re: benchmark of /dev/msg: it DOES slow disk access!
- Message-ID: <9208131284@umunk.GUN.de>
- Date: 14 Aug 92 05:30:39 GMT
- References: <3633222@zl2tnm.gen.nz>
- Organization: umunk (private)
- Lines: 27
- X-Newsreader: Tin 1.1 PL4
-
- don@zl2tnm.gen.nz (Don Stokes) writes:
- : hcp@csx.cciw.ca (H.C. Pulley) writes:
- : > So the original poster (Earthbound Misfit, I was it?) was correct: the active
- : > use of /dev/msg by mtalk DOES indeed reduce disk throughput.
- : >
- : > Any reason for this? (If you want the benchmak sourcecode and my numerical
- : > results, just ask.)
- : >
- : > Anyone know why mtalk should cause a fourfold cut in disk read time?
- :
- : This is just an educated guess, but doesn't the use of messages reduce the
- : available kernel space that can be used for the buffer cache? A reduced
- : buffer cache would certainly hit disk I/O performance.
-
- Really a well thought guess, but mtalk reduced the I/O performance also
- under Coherent 4.0. When you boot 4.0, it displays the size of the
- buffer cache, so I think I/O cache is either static or allocated at
- boot up. Also 4.0 hasn't the limit of a 64KB kernel pool, so I think the
- messages don't reduces the I/O cache. And the performance slows down,
- if mtalk is running without any input and for that, without any
- messages on their way.
- --
- Udo Munk | voice: +49 2131 275348 | uucp: +49 2131 278869
- Oberstr. 21 | mail : udo@umunk.GUN.de | 2400/1200/300 8N1
- 4040 Neuss 1 | ...!{mcshh,smurf,unido}!easix!umunk!udo | MYNAME=bbsuser
- Fed. Rep. of | CompuServe: 100021,2515 | login: uucp
- Germany | at first connect get ~/info and ~/files| pwd: public
-