*BSD News Article 6102


Return to BSD News archive

Path: sserve!manuel!munnari.oz.au!hp9000.csc.cuhk.hk!uakari.primate.wisc.edu!news.larc.nasa.gov!darwin.sura.net!spool.mu.edu!agate!doc.ic.ac.uk!uknet!mcsun!sunic!sics.se!jonas
From: jonas@sisu.se (Jonas Lagerblad)
Newsgroups: comp.unix.bsd
Subject: Hardware (cache) problem on 486DX-50.
Message-ID: <1992Oct6.110204.1100@sics.se>
Date: 6 Oct 92 11:02:04 GMT
Sender: news@sics.se
Organization: Swedish Institute of Systems Development
Lines: 44



I have been running BSD on two different brands of 486-50 boards, and had
some strange problems on both of them, which I have tracked down to 
the external cache memory. They are configured almost identical except
that one of them have 64k external cache and the other 256k. They both have:
    16 Mbyte memory
    Adaptec 1542B SCSI controller
    1 3 1/2" boot floppy
    1.0 Gb Wren7 SCSI disk.

I noticed the problems in two ways, one was that when the computer is
heavy loaded, random programs started to core dump and the other was that
the computers crashed when I was trying to run X386.

The first problem was reproduced by forever recompiling the kernel
(while 1 ; rm *.o ; make; end), this manifested itself most frequently on the
board with 256k cache and very seldom on the 64k cache board. The second
problem appeared with a 98% probability on both boards within 5 minutes after
starting X.

All these problems disappeared when I disabled the external cache.
When examining the cave memory chips I found that they were marked that they
have 20 ns access time (15 ns for the tag ram) on both of the cards.
One of the card manuals stated that this speed should be OK while the other
said that the cache memory should have 15/12 ns access time. At the moment I
assume that this is the problem, and run with only internal cache.

I have filed a complaint at the companies which we have bought the computers
from and I am eagerly waiting for the new memory, since it seems to have a
20% effect on performance to run without cache compared to having a 256K cache.

Has anyone else encountered problems like this, or am I the only one?

Does anyone know how the caches are designed (is it really possible to have 
the same access time on the cache ram as the clock cycle of the CPU?).

Regards 
Jonas Lagerblad
-- 
Jonas Lagerblad
Phone Office: +46-31-830250 Home: +46-31-205955 Car: +46-10-2846031
FAX:	+46-31-831047
email:	jonas@sisu.se