[lkml]   [2003]   [Aug]   [18]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [RFC] Re: Blender profiling-1 O16.2int
That makes sense, I was running a program that I found on IBM's website 
that's supposed to test context switching speed this weekend. It has 1
free lock and passes them around the group. If I put it up to 32
threads or so with one spare lock, I can start to see the starvation.
When running vmstat, it's apparent when the the starvation occurs as the
context switching sky-rockets. I was going to add to the example code
to check for how many times a thread wakes up waiting for the lock and
can't get it after reading that message about locks in the list. I
guess I won't have to do that now. Anyway, that'll bring my system to a
halt when the thread count gets up over 256. Still, it's usuable as
long as I'm not doing something else that makes heavy use of the processor.

I also had a nasty experience with a nautilus crash today that caused
this problem with X, I'm sure much similar to the results from Blender.
Obviously, X keeps getting expired because I was forced to reset after
waiting 1 1/2 hours for the system to shutdown (i.e. after the runlevel
changed to 6 to the actual rebooting of the system). Even after I
managed to get out of X (though I don't know if the process was able
ever to actually exit) the system took 5 minutes to get from "Stopping
sound services" to the next message. After that, I got a cursor that
didn't flash, then a blank screen but no reboot. I don't think X ever
exited since it never took me back to VT7, and the XDM shutdown comes
shortly before killing cron. The last message I saw indicated it was
shutting down service at... At that point, I could no longer change
between terminals though I could see continuing activity on my hard
drive. I waited 30 minutes from when I had exited X until I finally
reset. I had just clicked on the debug information button in bug buddy
when this started.

I think this problem is exacerbated when another app is competing for
the processor. The machine just pauses unless I'm also doing something
else, in this case compiling XINE. Once something is competing, it
looks like X takes an extraordinarily long time to come back into the
running queue.

Is there a way to figure out when a process is spinning on a wait and
exponentially decrease it's bonus if they are consecutive? I should
probably read through the source and some of these posts before I make
suggestions though, because I don't currently know much about how all
that works.


Con Kolivas wrote:

>Second, any applications that exhibit this should be fixed since it is a bug.
>Finally I still need to find a way to prevent this from transiently starving
>the system without undoing the interactive improvements to normal
>applications; I certainly don't intend to make them "work nicely" just for
>the sake of it.
>I do have some ideas about how to progress with this (some Mike has suggested
>to me previously), but so far most of them involve some detriment to the
>interactive performance on other apps. So I'm appealing to others for ideas.
>To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>the body of a message to
>More majordomo info at
>Please read the FAQ at

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:47    [W:0.132 / U:0.960 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site