* K.R. Foley <kr@xxxxxxxxxx> wrote:I will give that a try shortly.
I also have some other traces from this system that I have not seen
before on my slower system. For instance this one where we spend ~204
usec in __spin_lock_irqsave:
http://www.cybsft.com/testresults/2.6.9-rc1-bk12-S0/trace1.txt
Or this one where we spend ~203 usec in sched_clock. That just doesn't
seem possible.
http://www.cybsft.com/testresults/2.6.9-rc1-bk12-S0/trace4.txt
seems quite similar to Mark's IDE-DMA related hardware latencies. Does reducing the DMA mode via hdparm reduce these latencies?
Ingo