High traffic NFS performance and availability problems

[Borja Marcos]

> I think leaving the 4.x clients in a known configuration and just 
> varying
> the server configurations the right starting point.  Let's try tracking
> the server 5.x stability/performance first, then look into the client 
> 4.x
> crash reports.

	I've seen amazing performance differences between 4.9 and 5.3 in one 
case: a mail server with Cyrus-imapd and a MySQL database for 
authentication, using fiberchannel disk arrays.

	I couldn't do elaborate benchmarks (it is a production system) but I 
ran into a situation in which 4.9 spent about 90 % of CPU time in 
system (mostly syncer) and, just upgrading to 5.3 the problem 
disappeared.

	In both cases I had tuned some system variables (and I haven't changed 
the variables between 4.9 and 5.3). Notably, I set up a dirhash cache 
of 64 MB, because the machine had lots of huge directories.

	The workload is now the same, and it's working like a charm. The 
machine is a Dell Poweredge 2650 with two CPUs.




	Borja.