Package: iperf Version: 2.0.2-3 Severity: important I recenttly reported to the kernel folks that network performance as measured with iperf goes down significantly from 2.6.22 to 2.6.23 and I tracked it down to the introduction of CFS (see <http://lkml.org/lkml/2007/9/26/55>). During this discussion, Mike Galbraith noticed that iperf recently added a call to sched_yield() (see <http://lkml.org/lkml/2007/9/26/82>). Ingo Molnar concluded that this was bogus and suggested a proper patch for iperf: http://lkml.org/lkml/2007/9/26/132 Stephen Hemminger also claimed that a similar patch was submitted by the BSD people (<http://lkml.org/lkml/2007/9/26/209>, <http://lkml.org/lkml/2007/9/26/215>).
I suspect quite a few other users will see a slowdown as measured with iperf when they move to 2.6.23, so it would be good if iperf would stop using sched_yield() soon. Maybe you can apply one of the patches above (I tested Ingo's patch and it worked) or better yet work with upstream to get this into a new release. Thanks. -- Martin Michlmayr http://www.cyrius.com/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
