Subject: Re: New kinetic figures
To: Reinoud Zandijk <imago@kabel065011.kabel.utwente.nl>
From: Richard Earnshaw <rearnsha@buzzard.freeserve.co.uk>
List: port-arm32
Date: 02/08/2001 22:23:59
> On Thu, 8 Feb 2001, Richard Earnshaw wrote:
> > which takes
> >
> > Linux(x86/700MHz) real 0.3 user 0.02 sys 0.27
> >
> > Solaris(ultra5/360MHz) real 3.1 user 0.3 sys 2.3
> >
> > Shark/NetBSD(SA110/233MHz) real 9.2 user 0.02 sys 1.9
>
> (NetBSD times reported by `time ./a.out')
>
> Sun SPARCClassic 50MHz real 18.59 user 0.112 sys 3.641
> DEC Maxine 33MHz real 1:13.92 user 0.262 sys 36.582
> Pentium 120/160 MHz (?) real 3.21 user 0.022 sys 0.736
>
> RiscPC ARM7, 40 MHz real 26.42 user 0.114 sys 6.743
>
> so compared to these figures it only really tests pmap performance, not
> really processor performance.... You can see that the MIPS pmap code
> isn't performing that well too ... wich could explain the slow compiling
> speed too. The processor speed in user mode is more about what to expect
> from 33MHz compared to 50MHz.
>
pmap fork/exit performance is precisely what the test was intended to
cover. For shell scripts this is what dominates performance on ARM
platforms. Here are two more examples:
RiscPC SA110, 233 MHz real 30.2 user 0.06 sys 13.6
Pentium P90 real 5.69 user 0.02 sys 1.33
I know the memory system on a RPC sucks, but I'm sure we can do a lot
better than that.
> I would like to see a better benchmark ... hmm... what does
> /usr/pkgsrc/benchmarks say :)
>
> Cheers,
> Reinoud
>
>
>