[LAU] Worse performance in 3.4.x kernel over 2.6.33?

[Louis Gorenfeld]

Hi,
  I've been seeing noticeably worse performance using 3.4.5 and 3.4.11
PlanetCCRMA fc17 kernels over my old 2.6.33 PlanetCCRMA kernel, and I
was wondering if anyone else has experienced this. Running seven
instances of OP-X under WINE, for example, consumed about 10% more
CPU.

I also whipped out LatencyTOP to take some readings, and I see that
poll_schedule_timeout and pipe_wait seem to be culprits. What is
interesting is that the maximum milliseconds reading is staying
roughly the same for poll_schedule_timeout, yet the percentage
increases as much as 4x or more. I take this to mean that although the
worst-case is about the same between the new kernel and the old, that
on average it's waiting on this function longer.

I'm also running with a USB audio device, in case that might make a
difference. The realtime priorities, in general, look sane. Any advice
would be very much appreciated!

-Louis