Disk I/O and NFS issues

[Vidyasankar Narayanan]

Hi All,
I have been having some performance issues with one of my v2500's for a while. There is a considerable disk i/o issue and i noticed that some lv's are not striped and some disks in those lv's show up as 100% utilized. So i was planning to increase dbc_max_pct to 15% (currently at 5%) and also to stripe some of the lv's. Will this approach be effective? Also I'm planning to enable asynchronous i/o if this does not work. Are there anyone out there who run async i/o - any issues / problems?
My next issue is with the same server being an nfs server and one of the clients is very slow. The server is running 16nfsd's and the client is running 8 biod's. How much is the maximum no. of nfsd's and how do nfsd's and biod's correspond to each other (like if you run x biod's in client you should run 5x nfsds in server or something like this).

[CHRIS ANORUO]

There is a write up contributed by Berlene, search the forum with his name or use nfs as your search word.

[FRED Dennison]

Are the Disks connected internally or via a Fibre Channel Controller? We had a similar occurence where 'sar -d' showed 100% disk util. The cause was a Fibre Channel Controller on another host interfering with the operation of the Controller on the Host with the problem.

We have also experienced NFS problems, and were recommended the latest Patch Set (Search hpux - general forum for "NFS Memory Bottleneck"). Have yet to implement it though.

The NFS Performance Assessment document in the Technical Knowledge Base is pretty good for walking through the NFS System and troubleshooting it. Document ID KBAN00000261.

By running 'glance', you can also watch what a process is doing, and what is spends its time waiting for (s - select process, F6 - process syscalls & F2 - Wait States).

Hope this helps. FRED

[+¦+ó+±]

Hi,
If you have performance issue on NFS, you have to install NFS performance patch or use NFS verion 2. We use V2500 and 32 nfsd, 16 biod process. Also, you have to increase ninode in NFS server(In my case 36544).
And, if you use asynchronous I/O with oracle,
you have to check a patch related to async.
If you have PHKL_22126 or PHKL_22744, you have to run setprivgrp dba MLOCK(dba is oracle's group) at starting point.
Hope this helps.

[Dan Hetzel]

Hi,

NFS lack of performance is a known issue, solved by patch PHNE_22125.

Here is the link to the description:
http://europe-support2.external.hp.com/wpsl/bin/doc.pl/screen=wpslDisplayPatch/sid=c7b6cdb0119ff74f4c?PACH_NAM=PHNE_22125&HW=s800&OS=11.00

Best regards,

Dan