FreeBSD +NFS + mail services

Oliver Brandmueller ob at e-Gitt.NET
Tue Jan 3 22:59:28 PST 2006


Hello.

On Tue, Jan 03, 2006 at 08:05:02PM -0500, Francisco Reyes wrote:
> Looking for any links/comments/info on performance settings for a FreeBSD 
> mail setup using NFS.
> 
> In particular how does the client decides how many client nfs programs to 
> run? I noticed that 2 machines today were both using 22 client 
> connections each. Is that a hard limit in the FreeBSD nfs client?

Look at the sysctl vfs.nfs ->

vfs.nfs.iodmin: 4
vfs.nfs.iodmax: 20

In FreeBSD 6 you'll also find:

vfs.nfs.iodmaxidle: 120

Should be pretty self explanatory. How many you actually need depends 
very much on your configuration and how busy your servers are, how many 
mailboxes, how many mails per minute. Also it depends if your clients 
access e-Mails via IMAP or by POP3 (or whatever method apart from that). 
Especially IMAP can be very resource hungry.

So I cannot really give you hints on actual values, because they very
much depend on your setup. You should also play around with NFS read and
write data sizes, the bufpackets and such values until they best match
your setup with respect of speed of clients and server and the network
and again, how busy your setup is at the moment. Just run different
benchmarks in your test area or before going to production. And while
being in prod you could take the chance to make performance monitoring
on all the machines and carefully change different variables (i.e. use
of rdirplus, different buffer values). Also you could try to use Jumbo
Frames on your NFS backbone, given it is GigE only and _all_ of your
hardware and software (switches, every single NIC, every single
OS/driver involved) really supports this.

> Our current NFS server is a 5.4 machine and a new one coming online soon 
> will be 6.X stable. Client machines are 5.4

I just migrated our NFS server(s) from 5.4 to 6.0 - our clients are 
mainly still 5-STABLE, only few 6-STABLE clients. I saw a few NFS 
locking problems (lockd starting to spin) with the server while it was 
still RELENG_5, especially after stating with RELENG_6 clients. This was 
clearly load dependant (started usually on busy days when the backup was 
running into daytime). I haven't seen this yet with RELENG_6 servers, 
but the migration was just one week ago and the backup scheme had been 
changed also to avoid the problem, so I'm not completely sure here.

- Oliver

-- 
| Oliver Brandmueller | Offenbacher Str. 1  | Germany       D-14197 Berlin |
| Fon +49-172-3130856 | Fax +49-172-3145027 | WWW:   http://the.addict.de/ |
|               Ich bin das Internet. Sowahr ich Gott helfe.               |
| Eine gewerbliche Nutzung aller enthaltenen Adressen ist nicht gestattet! |
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-isp/attachments/20060104/cbe7a6e0/attachment.bin


More information about the freebsd-isp mailing list