I finally took a leap of faith and rebuilt part of the backup system. When I set it the first time around, I carved the MSA20 drive arrary into two logical RAID devices and in turn two logical drives on the Windows 2003 box it is directly attached to.
We use Backupexec 10 and I set up 36 or so B2D devices for each backup job, per day, per server to be backed up, all full backup jobs. I did this for ease of restore over efficiency. Due to the more limited space restrictions, I had to move B2D devices all over the place to get proper backups. It wasn’t working for me.
I blew the entire drive array away and made one big RAID5 logical drive just to build a bit of fault tolerance into the setup. I also created three B2D policies per server. Every other Friday, a full backup will run. The other Fridays, a separate full backup will run. Monday through Thursday, incremental backups will span the gaps between fulls. This way, I’m hoping to have two full weeks of hot backups. I’m going to add automatic duplicate jobs to copy the full backups to tape so I can have off site backups. We’re not a 24×7 shop, so backups during the weekend are mostly redundant. If space ends up not being an issue, I’ll add incremental backups for Saturday and Sunday as well just to cover everything. If I have a flaw in this design, please feel free to leave a comment tell me what I’m off on and what you might do differently.
Finally, Backupexec for Windows is not NCS(Novell Cluster Services) aware. So because of this, I have to make sure that I know where my services are located. If I have a resource attached to a specific node, I need to keep it there when it comes time for the backup to run otherwise it will fail. This throws a small wrench into everything. I really wanted to keep the nodes and resources completely non dependent on each other, but that’s not going to work out.