Solaris writing to Windows NFS Server - poor performance

Man of Honour
Joined
20 Sep 2006
Posts
35,573
I've got a Solaris 10 machine and I need to copy the Oracle DB backups to a Windows NFS share. The share is running on Windows Server 2008 R2. Read performance is limited by network speed as you'd expect, however I'm seeing a bottleneck of around 500 kb/s when writing.

Unfortunately installing any other apps on either box is out of the question as is using smb as the client built into Solaris 10 is terrible.

nfsstat -rc doesn't show any timouts or a high number of badcalls (<1%), the NFS log in Windows doesn't show much apart from lots and lots of 'write operation successful'.

nfsstat - shows nothing untoward.

Both are VMs and both are on the same network and ESXi 5.5 host.
 
Last edited:
1) Check NFS Synchronous writes on both ends. Make sure it's off / async writes.

2) Check volume write performance on the Windows box using another protocol - just to confirm underlying storage isn't the problem.
 
Check the mount options you're using in Solaris. Oracle recommend specific mount options for different purposes.

How are you writing the backups out - eg. directly to the NFS share through RMAN, or just a cold file copy?

My understanding was that NFS server in 2008 R2 was not very good and NFS was one of the significant improvements in 2012.
 
I've tried various mount options, none seem to have improved things so far. I've written a shell script to copy the files, so RMAN backs up the db and then I use a mv command to put them on the NFS share.
 
Yeah they are E1000, will try switching but I'm not sure that's the issue.

I created a smb share on the Solaris server and mounted it on Windows, saw 120+ MB/s transfer rate so I'm pretty convinced it's something to do with NFS performance.
 
Yes, same host. For testing I created a new 2008 R2 machine on the same host (to eliminate any physical networking issues). The host has local storage, where both VMs are stored.
 
Oh, just to add. I tried it to another server, a 2012R2 physical Hyper V host we have, same issue. So it's something on the Solaris end to do with NFS tuning I think.
 
Back
Top Bottom