Disk Utilization is very high

[Mahaveer]

Disk utils shows 100% utilization in the Glance tool. If I see the process Name under Glance which has high Disk IO Rate are nfsd.
there are 4 nfsd processes who had Disk IO Rate between 55/34 to 65/35

We are using this system as a NFS server also. Is this a NFS performance problem?
How can I reduce DISK Utilization or there is any other tool to drill down this problem

I have attached the snapshot of Glance

[Sundar_7]

Increase the number of biods you have running in the NFS clients. I believe biods will enable async IO. Refer the attached NFS server performance tuning guide. an excellent doc.

[Sundar_7]

Sorry the file is too big to be attached here.

http://docs.hp.com/hpux/onlinedocs/1435/NFSPerformanceTuninginHP-UX11.0and11iSystems.pdf

[Bill Hassell]

This isn't a problem as such. Someone (or lots of someones) are using your NFS server and it is trivial to massively load your server. For instance a remote user decides to find a lost file by searching every disk on the network. Using find (and du) on an NFS mountpoint will heavily load your server.

The disk I/O rate is actually low for typical disks. Since the NFS protocol isn't nearly as fast as your disks, you can probably improve performance a bit by increasing the number of nfsd processes in the file /etc/rc.config.d/nfsconf. There is a diminishing return as you add more nfsd processes. 16 is a good start, 32 might be a bit high and 64 may waste RAM and increase system overhead.

NFS is a very complex protocol to tune well and keep stable. Get a copy of Dave Olker's book on NFS performance for HP-UX: "Optimizing NFS Performance: Tuning and Troubleshooting NFS on HP-UX Systems" by Dave Olker

[sreejith_4]

Hi,
I also have faced the same issue. I have done a single change maxdsize increased to 1gb. And the problem was solved. I don't know what was the real issue.

Regards
Sreejith M

[Tim D Fulford]

I agree that it looks like nfsd may be the culprit.. but do check exactly which disks are busy (unless you only have one then it is trivial). Glance reports on the BUSIEST disk, so it could be that you are doing masses of swap to your root disk. do "glance -u" and make sure it is the disks that contain the FS being shared on NFS...

Regards

Tim

[rick jones]

You might check the mix of NFS operations your server is doing. If it is a lot of v2 writes, you may improvethings by getting clients to switch to NFS v3, which has a "sanctioned" way to have NFS writes "complete" simply once they are in the filesystem buffer cache. Then the filesystem can send them out to disc at a more leasurly pace and perhaps cut-down on the disc loading (particularly if the writes are to the same block(s)).

If you cannot get the clients to use NFS v3, then you might consider exporting the filesystem(s) with the HP-specific async option. However, only do that if you are satisfied with the satbility of your server.

And if the disc(s) are indeed saturated and glance isn't simply being a triffle paranoid, there is always the option of getting more and/or faster discs.

[Ted Buis]

If I remember correctly from Dave Olker's book, HP-UX 11i is better at NFS than 11.0, so that might be another option. His book is really good.