announce: v2.6.20-rt1 yum repository and kernel rpms
Curtis Doty
Curtis at GreenKey.net
Wed Feb 7 01:36:36 UTC 2007
1:41pm Ingo Molnar said:
>
> * Ingo Molnar <mingo at elte.hu> wrote:
>
> > $ ps aux | grep -i prio
> > root 31 0.0 0.0 0 0 ? D Feb05 0:00 [RCU Prio Booste]
> >
> > that artificially increases the system load average by +1.0. I've
> > fixed this in my tree and it will be in -rt3 today - it's harmless
> > otherwise.
>
> fyi, i've just released the -rt3 rpms - can you still see any weirdness?
>
It was -rt4 that came down the pike when I checked. And the /proc/loadavg
still seems decidedly "twitchy" for lack of a better term. It never fully
settles down. And a few shells running top and tload will eventually drive
it up to around 1 again. But it never stabilizes. This does not happen
with the old tickey kernels. And it does not happen in any deterministic
way I've seen before.
I notice all the irq threads are now -51 priority. And there are quite a
few of them. Plus the group of RT threads for each CPU. Your RCU Prio
Booster is only pri -50 and appears to be sleeping hapily. Are there
better profiling tools I could be running?
../C
More information about the fedora-devel-list
mailing list