[opensuse] NFS performance problem (lockd ?) in Suse 10.1
Hello. We have a NFS server which used to run Suse 9.1. Since it was upgraded to Suse 10.1, it is _very_ slow (nearly unusable for some things). In particular loging in with KDE in a client PC can take a long time. I have measured 5 minutes(!) from entering the password to the desktop appearing instead of the more usual 15-20 seconds. I now suspect that the problem is with locks since it doesn't seem to affect everything in the same way. Some info about our environment: - the server is a dual processor running kernel 2.6.16.27-0.9-smp in 64 bit mode. - the server has 4 SATA disks, configured in 2 RAID-1 arrays, each one with one ext3 file system. These file systems are the ones being exported. - the options in the /etc/exports file are (rw,sync) - there are some tens of clients with various versions of Suse Linux (from 10.0 to 10.2). In particular 2 classrooms with 16 computers each. When 16 students login at the same time, it takes minutes for kde to start (waiting for file locks ?). - the network between the server and many of the clients (including those 2 classrooms) is gigabit ethernet. - in a client: # cat /proc/mounts server:/fs /fs nfs rw,v3,rsize=32768,wsize=32768,hard,tcp,lock,addr=server 0 0 # netstat -in Kernel Interface table Iface MTU Met RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flg eth1 1500 0 173408 0 0 0 95072 0 0 0 BMNRU - in the server # netstat -in Kernel Interface table Iface MTU Met RX-OK RX-ERR RX-DRP RX-OVR TX-OK TX-ERR TX-DRP TX-OVR Flg eth0 1500 0514961025 0 0 0736413957 0 0 0 BMRU - so, there are no network errors, NFS v3 over tcp is being used and read and write sizes are 32KB. Two questions: 1 - Is someone using a Suse 10.1 as a NFS server without problems ? Do you have a comparable number of clients ? Are you using a SMP kernel ? 2 - Any suggestions about what could be the problem ? I tried several things to diagnose the problem: 1 - booting again in 9.1 solves the problem. This points strongly to some problem with 10.1, not a network problem or something outside the server. 2 - we had a problem with quotas not working in 9.1, so I tried to turn off quotas in 10.1, to see if that was the cause. No change. 3 - I wrote a simple script to test the NFS speed (just some reads and writes). There was some difference, but not dramatic (10.1 was maybe 1.5x slower than 9.1, not 10x slower). As I have several servers which are not currently very busy I setup some users that differ only in which server they are using (9.3, 10.1 or 10.2). With both 10.1 and 10.2 I was able to recreate the problem. Loging from 4 PCs at the same time it took 1:30 or 1:40 for KDE to start instead of 16 to 20 seconds it took using the 9.3 server. I have just repeated those tests and I got consistent results. The 4 clients PCs are running Suse 10.0. The options used in the mount are the same: s10.1:/home/s10.1/d1 /home/s10.1/d1 nfs rw,v3,rsize=32768,wsize=32768,hard,tcp,lock,addr=s10.1 0 0 s10.2:/home/s10.2/d1 /home/s10.2/d1 nfs rw,v3,rsize=32768,wsize=32768,hard,tcp,lock,addr=s10.2 0 0 s9.3:/home/s9.3/d1 /home/s9.3/d1 nfs rw,v3,rsize=32768,wsize=32768,hard,tcp,lock,addr=s9.3 0 0 Thanks in advance -- 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 ----- End forwarded message ----- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org
participants (1)
-
Rui Pedro Mendes Salgueiro