needs a tester with an SMP 7.0 box
m.rebele at web.de
Mon Feb 4 15:01:25 UTC 2008
Julian Elischer wrote:
> if you have a 7.0 SMP (8 way if possible) box and can stand
> a little down time, I'm looking for someone to test the following:
> This snippet may freeze the process after a while:
> with the following patch applied:
> it should not.
> can someone with a 7.0 machine please confirm this?
> (requires a recompile of libkse and linking with libkse.)
> (you could call it something else for the test).
I can confirm that the process with the applied patch runs far better than without - on a
7.0-PRERELEASE FreeBSD 7.0-PRERELEASE (i386). It's a P4 System with HT enabled (Kernel is compiled
Without the patch, the system is continuously accessing the Harddisk. The Datarate is about 6MB/s -
i don't know what's going on there. There's enough RAM marked as free (about 460MB), though no
reason for swapping.
I don't see this behaviour if the Code is linked against -pthread.
If i apply the patch to libkse, then the process linked against -lkse behaves like one against -pthread.
Without the patch, i've also the fatal error messages as stated by Mark Kirkwood, with the patch
they are gone.
Looks good to me!
More information about the freebsd-current