Re: NFS server slow in 10.1

Eric <ericteuber@xxxxxx> wrote:
Rui Pedro Mendes Salgueiro wrote:
Last week I upgraded our main server from Suse 9.1 to Suse 10.1 and
now the NFS server is sloooow (like, minutes to login to a KDE PC).

Was there some change in NFS that would explain this ? For instance a
change in defaults ?

Some informations about the server and client versions would be very

As I said, the server version is now Suse 10.1. The clients are mostly
10.1, 10.0 and some 9.3 . The network to most of the clients is gigabit.
But only the server changed.

helpful. How does your exports file look like and how are the shares
mounted on the clients?

exports: client(rw,sync)
client's fstab: defaults

Reading the man page, it seems that "defaults" is inefficient, and it
should be "rsize=8192,wsize=8192" . Also I could use async in the server,
but that is not recommended.

I will try "rsize=8192,wsize=8192" and see if it makes any difference,
but I doubt it will be dramatic.

I remember having the same problem a pretty long time ago. I think it
was related to the used nfs version which is 3 in 10.1 and was 2 in 9.1.
check man nfs and man nfsd to force a specific nfs version.

You mean that nfs v3 is orders of magnitude slower than nfs v2 ?!

Anyway I checked the differences between v3 and v2 (*) and v2 is limited
to 32 bits file offsets (so, maximum file size 2GB). I am almost sure that
I have not had that limitation for a long time.

* http://en.wikipedia.org/wiki/Network_File_System_%28Sun%29

I still have the 9.1 version in another partition in the server, so I
might boot with it and check which version was being used, but I doubt
that it was v2.

Currently it seems all of my clients are using v3:

# nfsstat -s
Server rpc stats:
calls badcalls badauth badclnt xdrcall
36112766 0 0 0 0
Server nfs v2:
null getattr setattr root lookup readlink
0 0% 0 0% 0 0% 0 0% 0 0% 0 0%
read wrcache write create remove rename
0 0% 0 0% 0 0% 0 0% 0 0% 0 0%
link symlink mkdir rmdir readdir fsstat
0 0% 0 0% 0 0% 0 0% 0 0% 0 0%

Server nfs v3:
null getattr setattr lookup access readlink
22 0% 8035938 22% 374222 1% 1057184 2% 1726647 4% 1881 0%
read write create mkdir symlink mknod
23754808 65% 720483 1% 103465 0% 787 0% 494 0% 24 0%
remove rmdir rename link readdir readdirplus
129389 0% 185 0% 27063 0% 55834 0% 203 0% 26721 0%
fsstat fsinfo pathconf commit
1063 0% 33 0% 2 0% 50846 0%


--
http://www.mat.uc.pt/~rps/

..pt is Portugal| `Whom the gods love die young'-Menander (342-292 BC)
Europe | Villeneuve 50-82, Toivonen 56-86, Senna 60-94
.

Relevant Pages

  • Re: Filesystem monitoring question
    ... server as a directory. ... A number of SMB and NFS clients, around 30, will need to write to this ... Writing to the directory is 24/7 since some clients VPN into the network ...
    (freebsd-hackers)
  • NFS UDP fast server slow client problems
    ... Fast NFS server causing SLOW NFS reads on Tru64 clients ...
    (Tru64-UNIX-Managers)
  • Re: NFS server slow in 10.1
    ... check man nfs and man nfsd to force a specific nfs version. ... I still have the 9.1 version in another partition in the server, ... Currently it seems all of my clients are using v3: ... remove rmdir rename link readdir readdirplus ...
    (alt.os.linux.suse)
  • Re: nfsrvstats.srvrpc_errs rapidly increasing
    ... > The srvrcp_errs are very likely unrelated to the hangs. ... > friends) from the NFS server routines in nfs_serv.c. ... other clients may behave strangely. ...
    (freebsd-net)
  • Re: FC8 and NFS service
    ... The NFS server is doing some kind of evil security check which was ... Since the port works against the FC1 server, ... Changing clients isn't going to happen, ...
    (Fedora)