NFS Write performance

[Freminlins]

On Wed, 02 Mar 2005 08:52:16 -0800, Tim Traver <tt-list at simplenet.com> wrote:
> Hi all,
> 
> ok, I've searched far and wide, but I have to ask the FreeBSD gurus
> about it...
> 
> I'm using a Netapp NFS server to serve up content to FreeBSD clients,
> and I am seeing terrible write performances.

I don't call myself a guru, but we have a similar setup with 4 NetApps
with about 20 FreeBSD (dual Xeon) clients.

> I've turned on these in the rc.conf file :
> 
> nfs_client_enable="YES"
> nfs_client_flags="-n 4"
> nfs_bufpackets=8

You don't need these for the client:
> nfs_server_enable="YES"
> rpc_lockd_enable="YES"
> rpc_statd_enable="YES"

> and I've got these in the sysctl.cnf file :
> 
> kern.maxfiles=32768
> net.inet.tcp.keepidle=3600
> net.inet.tcp.sendspace=65536
> net.inet.tcp.recvspace=65536
> net.inet.tcp.slowstart_flightsize=2
> kern.ipc.somaxconn=16384
> kern.ipc.shmall=65536
> kern.ipc.shmmax=268435456
> kern.ipc.nmbclusters=32768
> 
> I'm using 5.3-RELEASE on a dual AMD Opteron machine.
 
> I guess my question is, how do I make NFS writes fly ???
> 
> The reads seem to be pretty good. I know that the settings on the netapp
> are per their settings...

What sort of performance are you getting? On one of our already busy
mail servers, using just normal mounting on fast ethernet, I get:
 
bash-2.05b# time dd if=/dev/zero of=/mail/A-D/100M bs=1024k count=100
100+0 records in
100+0 records out
104857600 bytes transferred in 9.284085 secs (11294339 bytes/sec)

Not bad! We also get extremely good I/O on email, i.e. small files.

> Thanks,
> 
> Tim.

Frem.