Re: Thanks

Stuart Stock (stuart@gundaker.com)
Wed, 9 Apr 1997 18:14:54 -0500 (CDT)


On Wed, 9 Apr 1997, guy albertelli wrote:

> The Hp version is compiled for HP-PA Risc 1.0 using gcc -O3 (actually
> a modified version of gcc by Cygnus I believe)
>
> Compiling for specific targets has not improved performance when testing
> it on a variety of 9000/7X5 HP machines.
>
> Do you know what is the specific chip in that machine?

Well, the G50 is a 9000/S800 box running a PA-RISC 7100 96MHz with
seperate 256Kb data and instruction caches, but I found the problem: swap
thrashing. The box was spending all its time on disk IO and I neglected to
check free physical mem before running the timings.
After hours with a minimal load (and more free memory) 10 runs yielded
an steady 19 secs for the 2^22 timing benchmark. Not as good as a P90
(averaging 10 seconds), but much better than the 45-65 seconds I was
seeing before. A model E35 9000/S800 with a PA-RISC 7100LC 64Mhz with a
combined data/instruction cache of 256Kb runs the timing benchmark in 25
seconds average.

--
Stuart Stock				       stuart@gundaker.com
Systems/Security Administrator		       http://www.gundaker.com
Gundaker Realtors			       "Blowing a buffer as we speak"