Start a Conversation

Unsolved

This post is more than 5 years old

1079

April 4th, 2010 21:00

Backup of very large client grid filesystem not meeting SLA.

I have a IBM 3584 Library with 14 IBM 3592 type drives, server running Solaris 10 and 5 storage nodes also running Solaris 10 with the entire backup environment having 600+ clients doing a monthly backup of 600TB.

I have a large Windows client grid with more than 22TB of data divided in just 4 drive letters, the largest one being 14TB. All of them are on 4GBPS SAN. The backup data flow is on Ethernet. The grid is using HP Polyserve and has 6 nodes and backups are running through just one node although all drives(filesystems) are available on all nodes RW in mode. A Full backup of the system is taking around 8 days (and nights) today.

I am planning to re-architect this grid environment as its not meeting the SLA. I am going to segregate the windows drives to different grid nodes probably that will give me more bandwidth. Today also they are running simultaneously as I have client,server parallelism, target sessions etc to allow it going to two tape drives.

The summary is that I cant find a bottleneck anywhere. The average file size on the client is not so large, that increasing the bandwidth is going to help me it seems.

network is 1GBPS full duplex across the backup network(separate) in the grid clients with Networker server and storage nodes having etherchannel of 4GBPS. SAN is 4GBPS, performing just fine, IO latency on storage subsystem is pretty good. During backups, Grid node using only 5-10% of CPU, 5% memory, 15% network bandwidth. Legato server and storage nodes are just fine on memory,CPU,IO,n/w although loaded at peak time(nights) also due to other backups in the environment. Library performance is good, giving 30+Mb most of the time.

My problem is that even if I divide the drives to different grid nodes, and even through different storage nodes(DDS), backup is going to take more than 6 days because the largest drive is 14TB and cannot be reduced.

I am planning to redefine the file systems into multiple client resources and save set instances to have multiple streams of backup running simultaneously instead of just 4. Does this look like it is possible? Is there anything else that I can do? Any recommendations on LAN free backups?

I am running networker 7.4.5.

Thanks for going through my long mail.

76 Posts

April 13th, 2010 10:00

Hi,

     what kind of storage you are using in this grid? I guess powersnap could be the solution for you.

Regards,

Anderson Xavier

3 Posts

April 13th, 2010 11:00

The storage is IBM XIV behind IBM SVC

76 Posts

April 14th, 2010 11:00


I guess you can use NMM to backup your large filesystems thru VSS and make a LAN free backup environment.

14.3K Posts

May 3rd, 2010 06:00

I'm not familiar with IBM storage, but I assume you can do some sort of snapshots.  This would give you backup in fast way and you could simply mount on proxy where you would have you drives (or even keep it on storage if using small retention and having enough disk space).

No Events found!

Top