gcc test failures
bworthalex at yahoo.com
Tue Mar 15 18:09:26 PDT 2011
On Mon, 2011-03-14 at 12:57, Simon Geard wrote:
>Changing the priority doesn't really do anything to reduce load on the
>machine - it just means that those processes will be treated as less
>important when competing for CPU time. If there's nothing else running,
>there's no competition, so they'll still get all the CPU time they can.
>However, changing from -j2 to -j1 (or removing it entirely) may help,
>since only one process will be wanting CPU time at once.
I don't know a lot about linux scheduler, but I believe that it would make a
difference as, setting a lower priority means less time slice allocated to the
process thus, in turn reduces the frequency the process is scheduled. On the
other hand, increasing the priority would make the process scheduled every so
often (or perhaps for a bit longer duration depending on the exact
implementation) deferring the other lower priority processes until they can no
longer be deferred. Lowering the priority should help, in my opinion.
Last night I actually set the nice value for make to +20 and also used ionice to
set the priority to idle along with "-j1". That didn't make a difference. I kept
monitoring the cpu temperature. It never crossed 59C. Most of the time it stayed
between 50C and 56C. Despite of that, tests failed in exactly the same way as
I reran it with "-j2" and without changing the nice value nor the I/O priority
and the CPU temperature stayed between 50C and 59C. SO it seems that temperature
never crosses 59C even without the heat sink fan running (its -10C outside btw).
I don't think 59C would cause hardware faults as no other tests fail with the
cpu running at the same temperature range.
First, I am gonna try recompile manually to see if jhalfs setting some
unfavorable flags and thus making the test fail (I doubt that!). If that fails,
I'm gonna download and run prime95 to see how the hardware fares.
I will get back on this ASAP.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the lfs-support