From: sct@dcs.ed.ac.uk (Stephen Tweedie) Subject: Re: 486 caches -- testing and enabling Date: 11 Jun 1993 14:38:58 GMT
On 10 Jun 93 07:11:37 GMT, root@jacobs.jacobs.mn.org (Mike Horwath) said:
> : Fox> I was using the dhrystone2.1 tests. I have a DX2/66, so I would
> : Fox> expect to see a clear effect from both the on-chip and off chip
> : Fox> caches. The external cache is 256K.
> : The dhrystone test is so small it resides entirely within the 8KB on
> : chip cache. It will not make any use of the external cache.
> Under Linux (or any other *nice), running a small load wiil for sure flush
> the external (and internal I might add) cache because of context switches.
That won't help much. On my 486/33DX, a single dhrystone takes 35
microseconds. It does almost 300 dhrystones in 10 milliseconds, and
you only get a context switch every 10 milliseconds under Linux. In
other words, running dhrystone on a loaded machine will only cause the
cache to be flushed once every 300 complete iterations, so it will
have almost no effect at all on the measured performance.
Cheers,
Stephen.