[Gc] powerpc64 problems

skaller skaller at users.sourceforge.net
Wed Nov 23 00:09:57 PST 2005


On Tue, 2005-11-22 at 14:30 -0800, Boehm, Hans wrote:
>  Unfortunately, the sleep case arises
> fairly frequently, and Linux often can't sleep for less than 20msecs,
> which is painful.

FYI: linux scheduler on my kernel at least is woeful.

I ran 50 processes each doing CPU intensive calculations,
each was timed -- this is part of the performance measurement
mini-benchmark code in the Felix system (when any process
finished a new one was started).

For slower calculations, taking up to 60 seconds, one would
expect roughly 50 times the single process real time to elapse
on average, and the data to be bell curved fairly tighly around this.
That would show large numbers of fine time slices equally distributed
between the processes.

I hoped to see the advantages of various  compilers change, 
depending on whether the generated code ran under on
a loaded machine or not -- dedicated test environment is
fairly pointless, since performance depends on caching,
and context switches have to impact the cache.

What I actually found was the data scattered all over the place.
Some of the slower tests seem to hog the CPU, Linux never seemed
to schedule more than about 5 processes, and the min/max times
were spread over a huge range .. ;(

-- 
John Skaller <skaller at users dot sf dot net>
Felix, successor to C++: http://felix.sf.net



More information about the Gc mailing list