High CPU - kjournald & kblockd - nice level

Jim Cornette fc-cornette at sbcglobal.net
Tue Jun 22 21:39:25 UTC 2004


Brian Hanks wrote:

>Has anyone else experienced what I will call sporadic, jerky X response
>in FC2?  It seems that any time I have a background process running that
>is even remotely utilizing any CPU or IO, I have pathetic response in
>X.  
>
>My RH9 box and my FC1 box don't have this issue and they are slower
>machines with less memory and older disk.  As a test to rule out CPU or
>chipset specific issues, I've built a completely different machine with
>FC2.  Both machines exhibited the exact same behavior. 
>
>I quite reluctant to make the switch to FC2 without solving this issue. 
>It's quite annoying to have X become unstable every time I go to compile
>something in the background.  
>
>Hardware Setup #1:
>Asus A7V600 (VIA Chipset)
>1 GB RAM
>Athlon XP 3000+
>2 Samsung 120 GB SATA drives (RAID1 with LVM2)
>NVIDIA MX440
>FC2 fully updated
>
>Hardware Setup #2:
>Asus A7N266-VM (NVidia Chipset)
>1 GB RAM
>Athlon XP 2100+
>2 Maxtor 40 GB ATA drives (RAID1 with LVM2)
>NVIDIA MX400
>FC2 fully updated
>
>Upon further digging, it looks like kblockd & kjournald
>are chewing up a ton of CPU.  I'm even noticing it when I don't have
>large processes running.  It's definitely not as severe, but it seems to
>be occurring quite regularly. 
>
>The one test that I haven't done is to remove LVM2 and RAID1 from the
>picture.  I wonder if that would eliminate the problem?
>
>What exactly is the function of kblockd and kjournald?  Any insight
>would be greatly appreciated.
>
>Thanks,
>Brian Hanks
>bhanks at bhanks.com
>
>  
>
I'm not doing a lot of compiling on my computer. I recall something 
about the scheduler having priority levels that can be assigned to have 
processes use a lower or higher priority level.
 
I just tried to run
man nice
and it came up with an explanation.

My question about the completion of the compilation. Is the speed of the 
compilation quicker?

Jim





More information about the fedora-list mailing list