Kernal Parmeters for NFS cluster

[John Liden]

Hi All
I am setting up a 2 node cluster (SG) for our home directories, menu application and web reports. Any special suggestions on kernal parms to support approximately 4000 users. With maybe 1200 the max connections at any one time.

[Pramod_4 · Accepted solution]

Run sufficient number of nfsd's.
Good amount of buffer cache will be helpful.

Pramod

[Duncan Edmonstone]

John,

Read this document:

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

HTH

Duncan

[Ashwani Kashyap]

Use enough nfsd on the servers and enough biod on the client side .

Also while mounting on the client side use the right amount of timeout and rsize and wsize .

Usuallu rsize and wsize of 4096 and timeo of 20 is good enough .

[Bill Hassell]

Run, don't walk to the nearest bookstore or website to order the definitive book on HP-UX NFS by Dave Olker. You'll see that NFS is not a robust filesystem at all and can be clobbered by uncontrolled access on the LAN. You'll find a few kernel parameters like nfile but the key is to setup enough biod and nfs daemons in /etc/rc. And use nfsstat to monitor the activity.

[John Liden]

Thanks for all your good input. I will go off and do some reading now.