Polyserve Matrix FS 3.51 backup solutions

[kevin kramer_3]

I'm am now finding myself troubleshooting why my Tivoli TSM cannot scan and backup files any faster than it does. If I use the same node/server to backup the nodes local fs (non PSFS) I get great speeds. Using standard "rsync" works well also. What gives? I think these enterprise backup solutions don't know how to handle the PSFS. How are you backing up your Polyserve Linux cluster?

[macosta]

I cannot give a recommendation on how best to back up your data, especially when it comes to recommending a specific brand.

I will say that specific system calls are more expensive on a clustered filesystem than they are on a stand-alone filesystem. Calls such as stat() require a cluster-wide flush of the file to ensure accurate results. This means that rsync and incremental backups will be impacted, as they need to compare each and every file, meaning a lot of cache flushing.

This is also typically more pronounced on a filesystem with many small files as opposed to few large files, although it's apples-and-oranges between a clustered filesystem and a traditional single-node filesystem.

I do hope to hear back from others' on how they perform backups best. I'm curious to hear as well.

[Emil Velez]

There are alot of issues in the performance of backup software.

The previous author mentioned the stat system call which is quite a issue for polyserve. In addition if you have lots of small files in a single subdirectlry level 100K that will also not be optimal. Are the other nodes in the cluster reading and writing the same files while they are being backed up ?

We need some information on the types of files and access pattern to help. Remember that if your storage is EVA you could do a snapshot and mount the snapshot on a certain node and backup from the snapshot.

rsyncing the files to a temporary space so the files are not being accessed and backing them up from there is a reasonable option too. In fact the new 3.7 version of polyserve has a replication functionality between clusters feature that uses rsync.

Yes 3.7 is out.