[PlanetCCRMA] Re: fedora machine load

fonse006 fonse006@csusm.edu
Sun Jun 19 12:25:02 2005

>===== Original Message From Tracey Hytry <shakti@bayarea.net> =====
>On both mine and Alexia's machine we see the load thing also.
>We're both running complete FC3-planet setups with the 
2.6.11-0.15.rdt.rhfc3.ccrma kernel.  Alexia is running it on an athlon, and me 
on a athlon64.
>Both of our machines run real well with this kernel.  Both of us also have 
the nvidia drivers loaded so we can do online games, watch dvds, along with 
listening to and making music etc.
>I'm not really sure if the load is as high as the tools say they are, or are 
they getting fooled because they are being preempted by other things with this 
>PlanetCCRMA mailing list

I am nod one-hundred percent certain on this, but here is what I think 
happens.  If anyone knows for certain feel free to correct me.  How programs 
like top, time, and vmstat figure out the system load is based on how many 
processes are on the scheduler's process runnable que.  The problem that you 
guys are having seems to happen when you have a process that is in D state 
(interruptible sleep) for a long time. To find out if you have a process in D 
state try out this command- ps aux | grep " D ".  Anyway what happens is the 
process has a semaphore or a spinlock that another process needs to run.  That 
second process is still technically runnable so it stays in the process 
runnable que, but every time the os tries to run it all it does is check to 
see if it can get the semaphore or spinlock, decide it can't and then pass 
control back to the os.  The fact that this is happening, if in fact it is, 
should prolly be reported to the kernel developers if you can figure out 
exactly what processes are interacting to cause it.  For now it is not a big 
problem in the sense that  it is not very likely to degrade your system 
performance much because the runnable process never actually does any work 
other than a context switch. Anyway, I hope this helps.