Call for thread testers

Terry Lambert tlambert2 at mindspring.com
Thu Aug 28 02:25:39 PDT 2003


Daniel Eischen wrote:
> > Use a kernel without witness compiled in, libthr should be faster
> > than this result.
> 
> Well, what got me was the variance in time.  Most of the time
> it takes twice as long.  And it doesn't seem to hang for a few
> seconds at any one point; it plods along consistently, just
> twice as slow overall.  It could be mutex contention and
> false wakeups or something...
> 
> > But I always can not finish this test for libthr on my SMP machine,
> > in most time, it will deadlock, so I can not give you a reliable result.
> 
> It hangs in my thread yield test also; don't know why.

I never saw if Julian's IPI patch was committed or not.  I
would expect a lot of CPU wakeup latency without it, if you
HLT in the idle loop...

-- Terry


More information about the freebsd-threads mailing list