On Thu, May 28, 2009 at 5:12 PM, Deric Abel <[email protected]> wrote:

> We are running sles10 sp2 not redhat, so I am not sure if this bug was in any 
> of the red hat releases.  The kernel that comes with sp2 was 2.6.16.60-0.21, 
> and the most recent kernel for sp2 is 2.6.16.60-0.37  (remember that the 
> number after the "-" is the rpm build/release version, so that will be 
> different between distros).  I am not sure want the underlying problem was, 
> but from this mailing list it seems there was a bug in the sles10sp2 kernel 
> and it's fixed with the newest.  So unless someone else would like to explain 
> the bug, that's all the information I have.

The cause is that the application asks for a wakeup in 1 ms or so
(this is silly design since it is polling). The kernel only provides
10 ms granularity and rounded incorrectly to 0 ms, thus making the
silly thing pretty bad. It also shows very high CP overhead when this
happens.
The fix is that the kernel rounds it to 10 ms to limit the impact.

Martin just told me the fix is in: 2.6.18-124.el5 für RedHat RHEL5 und
2.6.16.60-0.34 für SuSE SLES10

-- 
Rob van der Heij
Velocity Software
http://www.velocitysoftware.com/

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to