online backups from another node in MC/SG cluster

[Mladen Despic]

I am looking into a possibility to use a 2nd mirror copy, then stop or freeze our applications, lvsplit the 2nd mirror, then restart the apps, then backup raw from the split mirror copy during the day. This seems to work fine on a single node. Is there a way to minimize the performance hit on the primary node by doing backup from another node, while the apps are running?
For some reason I thought I should be able to export the second mirror copy from one system and import it on another node, but then I realized the export/import works only with volume groups ?? Is there a way to do this? Any ideas?

[A. Clay Stephenson]

Hi Mladen,

I would look into the OnlineJFS snapshot mounts. The snapshots take only a few seconds and you can then resume the application. My backups then start immediately during the night when the loads are realtively low using the snapshot mounts. I find that the I/O hits are much smaller than the syncing that occurs when you re-mirror. I do suggest that if you try this, your snapshot buffers be mirrored. If the snapshot buffer fails, your main filesystem can hand but mirroring the snapshot raw lvols fixes that.

Clay

[melvyn burnard]

you are correct in the vgexport scenario, but there is no real neccesity to do this.
You could of course try to vgchange -a r on th ebackup node which activates the VG in read-only mode, and then mount the fs's in ro mode.
This has the problem that if the origional package/node failsd, the package will not start on this node.

I think the previous suggestion of OnLine JFS may be you best bet here, unless you had EMC or XP arrays, in which case you could use thing like the BCV split, and then vgchid on these to have a new VG to import on the backup node.

[Mladen Despic]

Clay:

Unfortunately, some of our lvols are raw (there is no file system), so I don't think I can use OnlineJFS Snapshot. Thanks for all the info though. It is quite educational to read your posts.

Melvyn:

Thanks for suggesting the read-only mode. It looks like it may be useful if we expand to a 3-node cluster (so that we can always have one node available for a package failover) ?

We WILL attach a cluster to XP256 eventually, so your second comment may be quite relevant. However, I'm not familiar with "BCV split". Any further pointers?

Thanks a bunch ... Mladen

[melvyn burnard · Accepted solution]

A BCV, or Business Continuity Volume, is a way within an EMC array whereby you can have a "mirrored" copy that can be split of as a new VG, and using the vgchid command you get what is to allintents an dpurposes a new VG, with areplica of the lvols' and from the origional that are now separated.
You then activate the VG on the other node, do your backup, finish it and then de-activate the VG, and remerge it on the origional node.

For the XP arrays, there is an equivalent but at the moment I cannot recal what this is called ;-}

[A. Clay Stephenson]

Hi again Mladen,

You might also look into VxVM ( the Veritas Volume Manager) which will make snapshots of logical volumes.
The BCV's work quite well.

More food for thought, Clay

[Carsten Krege]

The analogous product to EMC Timefinder (BCVs) for XP256/512 arrays is "HP SureStore E Business Copy XP", product number B9330A.


HP SureStore E Business Copy XP is an optional non-stop processing feature of the HP SureStore E Disk Array XP256/512 that provides the ability to nondisruptively create and control a copy of a source volume for multiple target volumes within the same disk array. Three target volumes, or copies, per source volume are permitted. The copies can be used by other applications or systems. This feature is useful for allowing non-production activities such as backup, data warehousing, decision support, testing, etc., on the copies.

For more info check

http://www.hp.com/products1/storage/disk_arrays/xpstoragesw/index.html

Carsten

[Mladen Despic]

Between "HP SureStore E Business Copy XP"
and "Veritas Volume Manager" I'm leaning towards VxVM as it is not tied to any particular disk array (??) However, given that we are going to use XP256, I would like to know if there are any advantages/disadvantages to go with the first option instead? Any info on this?
Thanks again... Mladen