nfs slow behaviour

Bob Marcan bob at interstudio.homeunix.net
Mon Jul 26 08:21:15 UTC 2004


John Thompson wrote:

> On Fri, 2004-07-16 at 14:54, Chadley Wilson wrote:
>
>>
>>> So I had working nfs and now I have a broken nfs, mmm.
>>>
>>> The server worked all night long no problems,
>>>
>>> My setup is as follows:
>>>
>>> Celeron 2.4 Gig with 512 mb ram on a msi 7005 maiboard on board
>>> everything,
>>> I have added thre pci intel pro 100 + network cards and fill the pci
>>> slots.
>>> I created a bond between the three intel cards (eth1 eth2 eth3). and 
>>> the
>>> onboard (eth0) is a route to another network ie eth0 = 196.25.100.120
>>> bond0 = 192.168.10.2
>>>
>>> When I start nfs it takes a long time time more than 5 five minutes.
>>> I have checked dmesg and /var/log/messages there are no entries 
>>> refering
>>> to nfs or nfsd. It was working perfectly until I added a new dir 
>>> which I have now
>>> removed, but I still sit with this problem.
>>> Exportfs does show the shares, but exportfs -r or -a just hangs for 
>>> more
>>> than five minutes too.
>>>
>>> I have tried rebooting and restarting each service one by one
>>> netfs
>>> nfs
>>> xinetd
>>> network
>>> etc...
>>> Where else could I look for the cause of the problem.
>>>
>>> I have tested the bond0 device thoroughly I just copied two gigs of 
>>> isos
>>> over in under 3 minutes, I don't think thats the problem.
>>>
>>> The service --status-all |grep nfs shows the service is running
>>>
>>> here is some output.
>>>
>>> [root at preload root]# service --status-all |grep nfs
>>> nfsd (pid 3210 3209 3208 3207 3206 3205 3204 3203) is running...
>>> [root at preload root]#
>>>
>>>
>>> [root at preload home]# exportfs
>>> /diskb/ifc1     <world>
>>> /diskb/ifc2     <world>
>>> /home/fc1       <world>
>>>
>>> Then try mounting and
>>>
>>> [root at chadlin root]# mount preload:/fc1 test/
>>> mount to NFS server 'preload' failed: server is down
>>>
>>> any ideas?
>>
>
> No solution, but I see a similar problem here using FC1.  My /home 
> directories are nfs-mounted from a FreeBSD (5.2.1) machine and as you 
> say it can take from 5 to 30 minutes for the mount to succeed -- when 
> it is done from /etc/fstab:
>
> amayatra:/home                 /home                   nfs rw 0 0
>
> I've actually commented out this line in /etc/fstab because it takes 
> so damn long to mount. No errors are displayed.  The nfs server shows 
> the mount request being received, but it takes fer-frickin'-ever to 
> complete on the client.  None of the other machines on the network 
> (running FreeBSD and NetBSD) have this problem.
>
> Curiously, if I leave the line commented out and mount /home manually 
> ("mount amayatra:/home /home") it is instantly mounted; no delay.
>
> I've tried changing when the nfs filesystems are mounted in init, but 
> it doesn't seem to matter.
>
> It's kind of a PITA because I have to mount /home manually when I 
> reboot, but I don't reboot all that often so I've just put up with it.
>
We have similar problem at one of the our customers. RH9, tg3 driver.
It takes too long to autonegotiate with the switch.
After that, everything works fine.
tg3 doesn't have any options, ethtool can only set 100Mb, mii-tool crashes.

Regards, Bob

-- 
 Bob Marcan, Consultant                mailto:bob.marcan at hermes-plus.si
 S&T Hermes Plus d.d.                  tel:           +386 (1) 5895-300
 Slandrova ul. 2                       fax:           +386 (1) 5895-202
 1231 Ljubljana - Crnuce, Slovenia     url:   http://www.hermes-plus.si






More information about the fedora-list mailing list