I’ve been trying to sort out backups in my client’s site over the last while. Part of the process has been to update Arcserve 2000 (7.0) server installations and agents up to Brightstor 11.5 SP2. We’ve had serious problems with the time it can take to run a full backup that starts on Saturday afternoon. For example, this last weekend’s backup finished after I arrived in this morning. We’ve also got the horrible issue of having to call in tapes to recover anything for a user … which is all too common.
The client had previously ordered a HP MSA 1500 and one of the guys attached it to the main Brightstor server last week. Today I installed the Disk Staging Option for Brightstor 11.5. I started mucking around to get to know it. There was zero documentation on the CA website to be found and the help file was less than helpful.
I quickly figured things out:
- You can create devices which are pretty much just folders where backups are stored on disk. I created one for full backups and one for differentials.
- I created a disk device group for full backups and again, one for differentials.
- I enabled staging on each of the disk device groups.
- I also made sure to increase the number of streams on the two disk device groups. Having X streams means that a Brightstor server can backup X agents at once. This is how to gain performance gains over tape backup. We’d noticed that agent-disk was no faster than agent-tape and we found no bottlenecks other than maybe the agent or agent machine itself.
- I enabled staging in both the full and differential backup jobs.
- I configured each job to use 10 streams, i.e. 10 servers will backup simultaneously to disk.
- I configured a policy in each of the jobs. A copy (migration) of data from disk to tape would commence 1 minute after the backup. Full backup data on disk will be automatically purged when it is 4 weeks old (the client runs a 4 week cycle plus monthly archives). Differential data is purged when it is 1 week old.
I tested this on two servers and streaming really made a difference. I could backup two servers in the same time it took to backup one. Having the data stream to tape afterwards took no time. In fact, disk-tape ran 10 times faster (LTO3) than agent-disk!
The first differential backup with disk staging will run tonight. I’m also running a perfmon job to capture statistics on the Brightstor server to see if 10 streams is too much or if maybe we can increase that without affecting server performance or introducing server-agent timeouts. We’re hoping that with 10 streams that we can get backup times to 20% of what they were.
What did I think of Disk Staging Option? I am no fan of CA (I’ve way too much experience with their "enterprise management" products and a constant need to call support and patch systems). I was pleasantly surprised to see how easy it was to configure. It also appears to offer some serious performance gains. However, it does appear to be very inefficient with disk. Each volume (or agent, e.g. SQL) creates a new backup file on the Brightstor server’s disk. This isn’t appended intelligently on the next backup. Instead, another file is just created. This does not compare well with Commvault’s usage of their "synthetic full backup" in Galaxy Backup. But, it is simple to set up and it appears to offer some serious reductions in backup times.