Re: Running mprime on dual-core machines

[Lars Nordlund <lars.nordlund%hem.utfors.se@localhost>]

On Fri, 13 Jun 2008 10:51:28 +0200
Thomas Klausner <wiz%NetBSD.org@localhost> wrote:

> Hi!
> 
> I recently imported math/mprime-bin, because IMHO it is a useful
> stability tester, since it stresses CPU+RAM together. In my
> experience, on a stable machine you can run it forever, but on an
> unstable one it dies within a few hours.
> 
> On a new -current dual-core machine I'm building, I ran two instances
> of it (because of the two cores). Usually one of the two dies within
> an hour (the other keeps running). A single one runs for more than a
> day without problems.
> 
> I would like to know if this is a problem in the hardware, kernel
> multithreading or linux emulation.
> 
> Could someone else with a spare stable dual-core machine on
> -current/amd64 (needs COMPAT_LINUX32) please run NUMBER_OF_CORES
> instances of mprime and report if it works for them?

One ran successfully all night and the other stopped after a little
more than one hour. Results.txt for that one said:

ERROR: ILLEGAL SUMOUT
Possible hardware failure, consult readme.txt file, restarting test.

Also seen during this; timekeeping is not good! It had gained 45
minutes during the night and after restarts of ntpdate-ntpd xclock is
now stuck. The second hand does not move unless I move a window ontop
forcing a refresh of the window. Hmm..


Best regards,
        Lars Nordlund