Re: XP guests sluggish after around 1 week of uptime

[Avi Kivity <avi@redhat.com>]

(re-add kvm@vger)

jmandawg wrote:
>> -----Original Message-----
>> From: Avi Kivity [mailto:avi@redhat.com]
>> Sent: Monday, January 19, 2009 7:06 AM
>> To: jmandawg
>> Cc: kvm@vger.kernel.org
>> Subject: Re: XP guests sluggish after around 1 week of uptime
>>
>> There is a lot of wait time (last column, wrapped around to the first in
>> your output).  Not a lot of real I/O though.  Can you describe your I/O
>> subsystem?  Are you using NFS?
>>
>>     
>
>   
>> What does 'top' show for the guests?  I'm interested in qemu's VIRT and RES
>>     
> columns.
>   
>> Also, 'vmstat 1' on host and guest while slow copying is happening.
>>     
>
>
> I'm using EXT3 on LVM on top of RAID 5.  I can try moving one of the guest
> images to a different disk without the RAID 5 if you think that would help.
>   

Software RAID or hardware RAID?

I think the root cause is the RAID here, not kvm.

Some suggestions:
- use the raw format directly on top of LVM.  That is, use -drive 
file=/dev/volgroup/logvol.
- if you use the raw format, disable disk caching: -drive 
file=/dev/volgroup/logvol,cache=off
- try non-RAID5 storage

To migrate your volumes, use

  qemu-img convert /path/to/image.qcow -O raw /dev/volgroup/logvol

Be sure to size the volume appropriately (same size as the virtual image).
> I just restarted the host machine yesterday, so the file copies are not
> running as slow as before, but still slow.
> Here is the current output of top during a file copy:
>
> top - 07:31:53 up 18:09,  4 users,  load average: 0.74, 0.57, 0.46
> Tasks: 157 total,   1 running, 156 sleeping,   0 stopped,   0 zombie
> Cpu(s):  6.7%us, 12.7%sy,  0.0%ni, 59.6%id, 19.9%wa,  0.0%hi,  1.0%si,
> 0.0%st
> Mem:   8190236k total,  8141100k used,    49136k free,  1246732k buffers
> Swap:  8388600k total,      228k used,  8388372k free,  2418596k cached
>
>   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
>  5945 root      20   0 2131m 2.0g 1924 D   13 25.8  44:15.71 kvm
>   

S=D: waiting for disk.  Low cpu usage as expected.

>
> Here is the current output of vmstat 1 during a file copy:
> procs -----------memory---------- ---swap-- -----io---- -system--
> ----cpu----
>  r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id
> wa
>  0  1    228  46008 1210608 2466132    0    0  2308  8024 2695 13290  7 17
> 63 13
>   

This seems fine if a bit low.  You'll get better numbers sitting 
directly atop LVM.

-- 
error compiling committee.c: too many arguments to function