On Tue, Aug 18, 2009 at 3:50 PM, john stultz<johnstul@us.ibm.com> wrote:
> On Tue, Aug 18, 2009 at 3:31 PM, john stultz<johnstul@us.ibm.com> wrote:
>> I'm seeing some bad scheduler behavior with this patch. I've seen
>> occasional bad behavior with 2.6.29-rt as well, but with .31-rt its
>> much worse.
>>
>> I'm running a SCHED_OTHER cpubound benchmark  (censored to
>> xxxxxxxxxxxxx in the logs) that runs for a number of minutes. With
>> 2.6.31-rc6 I'm seeing really poor numbers and if I cat
>> /proc/sched_debug I see all of the cpu bound tasks are  running on the
>> same cpu, while the other procs are basically idle. See the log below:
>
> Thomas didn't like that I censored the benchmark, so until I know
> better about what I can say there, I've reproduced a similar effect
> with a make -j8 bzImage. See below:

Same thing happens w/ 2.6.31-rc6-rt4. config attached.

Simple busy loops like:
while true; do ls > /dev/null; done &
while true; do ls > /dev/null; done &
while true; do ls > /dev/null; done &
while true; do ls > /dev/null; done &

Also show the same behavior.

thanks
-john