On Fri, 29 Jan 1999, Neil Conway wrote:
> Rik van Riel wrote:
> >
> > On Fri, 29 Jan 1999, Neil Conway wrote:
> >
> > > The behaviour is equally likely to change if I run the code under bash
> > > instead of tcsh - sometimes bash runs it quicker (and REPEATABLY QUICKER
> > > I mean, not once) and sometimes tcsh runs it quicker. The behaviour
> > > stays the same for a while until something in the innards of the machine
> > > (memory-mapping?) changes and then the speed-changing can either invert
> > > polarity or disappear for a while... Quite often when it disappears I'm
> > > left with repeated runs giving me the LOWER speed which is a shame.
> >
> > As Richard said, it's most probably cache line aliasing.
> >
> > This severely bites me too. On my UP Pentium 100 w/
> > Triton chipset x11amp used to use 35 to 40% of the
> > CPU. On my dual P120 w/ Neptune chipset the CPU usage
> > varies between 60 and 95%.
> >
> > The huge increase probably comes from the facts that
> > x11amp is switching CPU too often and the more
> > primitive L2 cache mapping on Neptune boards.
>
> Well CPU switching doesn't explain my observations as I have lots of
> instances of seeing it on UP machines.
>
> Dunno about the cache issues - to be honest I don't know why different
> page layouts in physical RAM makes such a big difference...
>
direct mapping maps many physical pages into the same group of cache
lines, _if_ you get an unfortunate layout in physical memory... If you
are lucky, it runs fast, unlucky it runs slow... but the 'variation'
kills me when I am trying to 'optimize'...
> Neil
>
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/mentre/smp-faq/
To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
