RELENG_7 jerky mouse and skipping sound (still a problem -BETA3)
joe at skyrush.com
Mon Jan 14 09:13:41 PST 2008
On 1 Jan, 14:17, Kris Kennaway <k... at FreeBSD.org> wrote:
> > OK, can you obtain a schedgraph trace when the problem is manifesting?
> > See /usr/src/tools/sched/ and previous discussion in this or related
> > threads.
I just recently installed 7.0-RC1, and I am seeing pretty severe "mouse
jerkiness" or "mouse freezing" while, e.g., compiling (as others have reported
here). It's not just the mouse, but keyboard events are also delayed in
the same manner (seen by holding down a key in xterm, e.g.). I am on a UP
2.4GHz P4, using PS/2 mouse (with moused) and keyboard.
I'm glad I found this thread, since you are asking for traces. I really hope
my traces help; this problem does seem like a regression from 6.2 (I had seen
slight mouse non-interactivity there too at times, but not nearly as bad).
Also, with Linux's new CFS making mouse movement *very* responsive, I think
it's vital that FreeBSD address this to avoid such comparisons.
I have tried both SCHED_4BSD and SCHED_ULE. 4BSD is a lot worse when
compiling, say, the kernel. ULE is better when compiling, but still has
issues with, e.g., firefox loading a page, catching up on multiple xterm
window resizing (see below), etc.
This trace is while using SCHED_4BSD and compiling the kernel / moving mouse:
And here are three traces using SCHED_ULE:
Please check out all three, in case I did not get a good sampling of mouse
events and compiles in any one...
Strangely, ULE exhibits mouse jerkiness more than 4BSD for the following: I
opened an xterm and dragged the right edge of the window back and forth
quickly, making the window wider/narrower. It is obvious in FreeBSD that this
queues up events for X (after some time, the window border no longer follows
the mouse at all), and if I release the mouse button at that time, leaving the
window narrow and immediately move the mouse in circles, it is jerky for a
while, then returns to smooth action after about 5 or 10 seconds. 4BSD is not
as severe in this one case, and I never see this at all in Linux with CFS
(i.e. kernel 2.6.23) - the window resizing never really gets behind like this.
Here is a trace showing this for ULE (xterm still catching up, if I
remember correctly, at end):
Here is one for 4BSD (xterm caught up before trace stopped):
As an aside, renicing Xorg and moused to -10 seems to help smooth the
mouse when using 4BSD when compiling, whereas it is not needed (and
seems to have little or no effect) when using ULE (even though, as I said, ULE
still shows jerkiness).
More information about the freebsd-stable