Re: 2.2.15 slow NFS

From: Jens Benecke (jens@pinguin.conetix.de)
Date: Wed Jun 07 2000 - 03:15:24 EST


On Wed, Jun 07, 2000 at 10:43:42AM +1000, Neil Brown wrote:

> On Wednesday June 7, jens@pinguin.conetix.de wrote:
> > Hi,
> > please CC: me as I don't read linux-kernel regularly.
> Who has the time!!

That's exactly the problem ... I started reading it but then it grew and
grew and I simply couldn't find the time any more. I am reading the digest
versions available at kernelnotes.org and such, its not the real thing but
it's ok.

Thanks for answering. :)
 
> > I'm running Mandrake Linux 7.0 (2.2.14-15mdk) on a NFS client, and Debian
> > 2.2 (but with a Mandrake 2.2.15-6mdk kernel) on a NFS server. NFS speed from
> > server to client is satisfactory (about 4MB/sec over 100MBit Ethernet). The
> > other direction is really, really slow. (100k/sec)
> Writing over NFS is usually slower than reading, though a factor of 40
> does seem a bit steep.
 
> 1/ What flags are you using when exporting. In particular, do you
> include the "sync" flag? You should for safety, but there is a
> speed cost.

# /etc/exports: the access control list for filesystems which may be exported
# to NFS clients. See exports(5).

/home 192.168.1.10(ro) 192.168.1.101(rw) 192.168.1.100(rw)
/home/ARCHIV 192.168.1.10(ro) 192.168.1.101(rw) 192.168.1.100(rw)

on the client:

server:/home on /home type nfs
(rw,nosuid,nodev,bg,rsize=8192,wsize=8192,hard,intr,nolock,addr=192.168.1.100)
server:/home/ARCHIV on /home/ARCHIV type nfs
(rw,nosuid,nodev,bg,rsize=8192,wsize=8192,hard,intr,nolock,addr=192.168.1.100)

> If you are using "sync" try using "async" and do the same tests.
> That might help isolate where the slowness is.

Another interesting thing is that when copying, the xosview network meter
jumps between 2MB/sec and 0MB/sec all the time, about 2-3 jumps/second
(both on server and client, reading/writing resp.). So it's probably not
general slowness, but "jumpiness" that's causing the problem.
 
> 2/ What size files are you writing to? If there are big enough to
> have double-indirect blocks, and you are using "sync", then the
> implementation of fsync in 2.2 hits you REALLY HARD.

I noticed the problem with some RPMs, but mostly WAV files, 5..200MB in
size. I was wondering about this because I didn't have this problem before.
(I just upgraded the server to 2.2.15 from 2.2.14, did anything significant
change between those two?)

> Compare writing to large and small files with and without sync, and
> let me know what you get.

Thanks again for answering. I'll try to pinpoint the problem and then mail
you back.

-- 
Microsoft is a cross between The Borg and the Ferengi.  Unfortunately they
use Borg to do their marketing and Ferengi to do their programming. 
                                        -- Simon Slavin, in the Monastery.


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



This archive was generated by hypermail 2b29 : Wed Jun 07 2000 - 21:00:27 EST