Re: Timings for optimised poll(2)
Richard Gooch (rgooch@atnf.CSIRO.AU)
Sun, 24 Aug 1997 11:27:02 +1000
Peter Leif Rasmussen writes:
> Hi Richard,
>
> You write:
> >
> >I've appended my latest version test programme. Compile with -O2. This
> >contains a *realistic* scanning loop. It takes 1.5 milliseconds on a
> >Pentium 100.
> >
> And just for fun I compiled it and ran it on a 486DX2/66MHz, a 486SLC2/50MHz
> and a 386SX/20MHz, and my numbers said:
>
> 486DX2/66MHz: 4200 us (linux 2.0.30)
> 486SLC2/50MHz: 10000 us (linux 2.1.51)
> 386SX/20MHz: 24400 us (linux 2.0.29)
>
> Without having access to any Pentium computer, I find it difficult to believe
> that you got 1.5 us with your test program?! If you did, could you
> give a point of view as to why you think the numbers are so different?
The same test takes 342 microseconds on a PPro 180 MHz. Why the
difference? Because the Pentium is a better chip than the 486 and
because the memory bandwidth is better. Similarly, the PPro is better
than the Pentium and PPro systems have a higher bandwidth to L1 cache,
L2 cache and main memory than Pentium systems.
Regards,
Richard....