On Tue, 2005-03-22 at 09:22 +0100, Ingo Molnar wrote: > hm, weird, and i have no solution for this yet. But i just found a > related bug in the -RT patch in that it reverted a latency breaker in > the ext3 path that your trace shows - affecting PREEMPT_DESKTOP. Could > you try the 40-03 patch i just uploaded (assuming it's stable for you > otherwise ...) and see whether you can still reproduce these latencies? > (if yes then please also send me a full trace via private mail, or bzip > -9 it in your public mail.)
Ingo, The problem is fixed. Now the longest latency I see with "dbench 16" and PREEMPT_DESKTOP is 591us in the ext3 reservation code. Trace is attached (compressed) in case anyone is interested. But I do not consider anything under a millisecond to be a problem with PREEMPT_DESKTOP. Lee
latency_trace.bz2
Description: application/bzip
