lkml.org 
[lkml]   [2002]   [Feb]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: kernel nfsd consuming 100% CPU on 2.4.17 and 2.4.18 with reiserfs?
Chris Mason wrote:
>
> On Tuesday, February 26, 2002 08:48:02 AM -0500 Brian Ristuccia
> <bristucc@sw.starentnetworks.com> wrote:
>
>
>>It seems that kernel nfsd consumes an inordinate amount of CPU time
>>during writes on this machine. With a few hundred kb/sec being written
>>over NFSv3 from a 2.2.17 client, all of the nfsd threads each consume as
>>much of the available CPU time as possible. On a similarly configured
>>machine with ext3 instead of reiserfs, nfsd consumes much less CPU time.
>>
>>Is there a known issue with NFSv3 performance and reiserfs?
>>
>
> No, it is not a known issue. Does it only happen with a 2.2.17 client, or
> can you reproduce with any kernel version on the client?
>

I can get it to happen with 2.2.19 and 2.4.4-pre3 as well.

So I'm pretty sure the NFS server is doing too much work somewhere.

If it matters, it's a SMP kernel running on a dual 1ghz pIII system with
2gb of memory. The filesystem resides on a linux kernel md RAID-5 array
with 6 10,000 rpm disks. It's my understanding that the a machine this
large should soak out the available network or disk bandwidth long
before it became CPU bound serving NFS. I also did some raw IO tests to
confirm that the md block device wasn't hogging up CPU time that was
getting accounted to the nfsd kernel threads. I can soak that array
pretty hard without soaking the CPU.

The closest machine configuration wise that I have access to is
similarly configured, only with 3 disks instead of 6 and ext3 instead of
reiserfs. Both machines were running exactly the same 2.4.17 image when
I started having this problem. I can't reproduce the problem there, even
when I do nasty things like run bonnie++ over NFS. (This isn't to say
that nfsd is free on this other machine, but I'm seeing it use on the
order of 2-4% CPU per nfsd thread with 8 threads and a load average of
between 1 and 2 vs. 20+% and a load average of 8 on the other machine).

Thanks.

--
Brian Ristuccia


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:20    [W:0.421 / U:0.044 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site