Igor Schein on Tue, 1 Oct 2002 17:54:27 -0400

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

pari as smp benchmark


timing of the following command can be used as a decent benchmark of a

echo ';'|gp -qf -p `echo 2^31-1 | bc`

Using this command, I discovered that SMP performance on Linux sucks.
On a dual CPU 1GHz PIII PC running redhat SMP kernel, the execution
time is twice slower if I start this command simultaneously on both
CPUs.  On a 2.4Ghz Xeon PC, it's 5(!) times slower.  Needless to say,
this smp performance degradation doesn't exist on every other platform
I tried.

I would like to report this problem to linux kernel mailing list.  I
am not 100% sure though how to describe this task computationally - is
it L2 cache-intensive ( my guess ), fp-intensive, etc. ?  I've heard
stories of kernel people being difficult, so I would like to present
this problem in the most coincise fashion, so that it's not ignored.
I believe pari users will also benefit if the kernel is fixed. 

So if any is willing to describe in simple terms the nature of this
task, I'd greatly appreciate it.