a nfs configuration scalability question

OK..
first of all my apolagies if this is a *developer* only list.

I'm running apache2 on rh-kernel on a quad processor machine.
the docroot (where it grabs the HTML files) is being served off a netapp750 filer.

what I'm finding is that under *any* load the runqueue goes up to ~700 and the cpu shows 100% idle. vmstat shows that all the procs are in 'blocked' state.

now.. I've tweaked the nfs mount a tiny bit to make it read only, and have a rsize of 8k, but it didn't seem to make any difference.

$ mount
cn-sfo1-pma-filer1:/docroot on /var/httpd/online/docroot type nfs (ro,rsize=8192,nolock,addr=10.10.112.27,addr=10.10.112.27)

the actual benchmark involves a HTTP client requesting a random URL which reads in 10 files (9 of which are the same for each request)

on Solaris (where it is currenlty running) I can get ~1000 nfs ops/second (mainly lookups)

so my questions:

* what tools are there for me to monitor WTF is going on (ie.. cache hit rates inode hit rates)
* can I tweak the cache/inode size
* are there tweaks/kernel patches which will help client performance for my type of workload
* am I being a complete dunderhead and missing something completely obivous (if so where should I RTFM ?)

Thanks
this nfs problem is a major roadblock for me, and would appreciate ANY pointers

Ian Holsman

-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs