NFS issues with remote mount during data protector backup.

[David Eichberg]

Hi,

We have four L2000 HP9000 servers running HP-UX 11.00.
One of the servers (cimdev) has an exported directory that is remote nfs mounted on the other 3 servers, which each run an application that uses Oracle 8.0.6.
Whenever we run a backup (using Data Protector) on cimdev, the other 3 servers experience nfs issues.
The syslog entry on each server logs errors as follows:

"Mar 23 12:33:38 glaxotab vmunix: NFS server cimdev not responding still trying"

When this problem occurs, the application running on the three servers basically freezes as Oracle halts. There are no errors recorded in the oracle alert log files, but it is completely locked. We can't even initiate an sqlplus session.

The remote mount holds a mirror of the oracle redo log files.

If we kill the backup on cimdev, the application resumes and runs normally.

This issue has only occurred recently, after we re-booted the cimdev server.

Prior to that, the backups worked fine without affecting the other servers.

Initially, we thought that there was some kind of locking occurring when backing up the exported directory, but we've actually excluded the exported directory from the backup script and the problem still occurs.

We could drop the remote mount as it isn't critically necessary, however I'd like to avoid that if possible.

I'm assuming that data protector is somehow affecting nfs but I'm not sure how.

Is anyone aware of nfs issues and data protector? (or have any suggestions on troubleshooting this).

Many thanks,

Dave

[Steven E. Protter]

I hope you are not running any oracle data across NFS because that setup is simply not supported.

The condition you are running into happens when you have a hard nfs mount and network communicaiton is lost between the client and th e server.

It could be that the server goes down. It could simply be network congestion prevents packet trasnfer long enough for the connection to go stale.

A couple of possible things to be done:

Do the NFS connection over a private network or VLAN segment. The vlan segment, using a bridge or a router prevents servers that don't need to be on the same collision domain from being there and causing congetsion.

The private network is even better. Its physically seperated and the only machines on that network are those that need to be there. We have a private network for Ignite and that can be used.

Its relatively important to keep windows clients off the same network as the NFS connection. The reason is those clients and servers are notorious for flooding networks with worthless, fragmented packets taht can congest and degrade NFS connections.

The log file you want to look at to track this problem is /var/adm/syslog/syslog.log

SEP