amd64/147926: system freezes due to heavy io to the disks

[Garrett Cooper]

On Sat, Jun 19, 2010 at 1:00 AM, Andriy Gapon <avg at icyb.net.ua> wrote:
> The following reply was made to PR amd64/147926; it has been noted by GNATS.
>
> From: Andriy Gapon <avg at icyb.net.ua>
> To: Dominic Duklas <dodo1122 at gmail.com>
> Cc: bug-followup at FreeBSD.org
> Subject: Re: amd64/147926: system freezes due to heavy io to the disks
> Date: Sat, 19 Jun 2010 10:55:22 +0300
>
>  on 17/06/2010 03:36 Dominic Duklas said the following:
>  >> Number:         147926
>  >> Category:       amd64
>  >> Synopsis:       system freezes due to heavy io to the disks
>  >> Confidential:   no
>  >> Severity:       non-critical
>  >> Priority:       low
>  >> Responsible:    freebsd-amd64
>  >> State:          open
>  >> Quarter:
>  >> Keywords:
>  >> Date-Required:
>  >> Class:          sw-bug
>  >> Submitter-Id:   current-users
>  >> Arrival-Date:   Thu Jun 17 00:40:00 UTC 2010
>  >> Closed-Date:
>  >> Last-Modified:
>  >> Originator:     Dominic Duklas
>  >> Release:        9.0-CURRENT
>  >> Organization:
>  >> Environment:
>  > FreeBSD lain 9.0-CURRENT FreeBSD 9.0-CURRENT #2 r209225: Wed Jun 16 01:47:01 BST 2010 root at lain:/usr/obj/usr/src/sys/LAIN amd64
>  >
>  >> Description:
>  > I'm currently running freebsd on my desktop (core i7 920, 6gb of ram). Since around January I'm having problems with the system freezing for short periods of time and eventually locking up due to heavy reads/writes to the disk.
>  > The cursor would freeze for a split second and songs in my music player (quodlibet) would skip completely (the error logs say that the music player was unable to access the song which was currently playing, even though my music collection is on a completely different hard drive than the os).
>  > After 10-20 minutes of the occasional freezes (the frequency is random, sometimes it can go without any freeze for some minutes or it can freeze many times in a short period of time) the system locks up completely and reboots itself. There is no crash dump produced by the kernel, nor anything shows in dmesg when the short freezes occur.
>  > I tried putting freebsd on a different hard drive, changing filesystems from zfs which I currently use to ufs but the issue doesn't go away.
>  > Today I tried using the geom rr sched which was recently committed to current to see whether it would make a difference, but it didn't help.
>  > This issue is not related to swapping as ram usage never gets high enough, and is also not directly caused by heavy cpu usage, although it might be part of the problem.
>  > I've done several clean installs of freebsd current since the issue first appeared.
>  >> How-To-Repeat:
>  > make -j8 buildworld would trigger it in my case, or 2 simultaneous while loops with extracting a large file archive or compiling something big like openoffice/gcc/boost from ports.
>  >> Fix:
>
>  You haven't let us know some important details.
>  What disk controller do you have?
>  What driver do you use for it (if there are multiple choices)?
>
>  P.S. Why did you file this bug under amd64 category?  Have you tested how i386
>  version of FreeBSD works on the same machine and came to conclusion that it's
>  only amd64 version that is affected?

Disk controllers would be helpful, but a full kldstat -v would suffice
as there might be some other GIANT locked drivers (like nvidia-driver)
causing issues. I say this because I do see a sporadic performance hit
with concurrent accesses via UFS2+SU when transferring files and
compiling applications using an ada(4) enabled setup, plus mfi(4), but
that is apart from the issue, but that is primarily because UFS2 is
giant locked within itself and the fact that there's other contention
arising from me running GUI applications like Firefox on my machine.

Andriy is correct though... we need more details as to what you're
running on the machine driver wise, your machine's stats, and/or a
mean number of running processes, threads, etc. If we don't have that
kind of data we can't possibly triage the issue and make a
recommendation on how you should improve your machine's hardware
topology and/or software config.

Thanks,
-Garrett