I currently preside over perhaps the most over-worked backup server in history (so much so that approx 50% of backups fail every night). To give a little more background, the "new" server took the place of the "old" one that was running BackupExec and hooked up to a VTL that was pretty much failed. This has resulted in the new NetBackup solution becoming swamped in very quick time as it was never really built or configured for the amount of data that it handles at the moment (I inherited it the whole shebang).
I have done a lot of reading round the subject over the last few months, so I have a reasonable idea of what I want to do and why and backed it up with some solid math. The trouble is that with the economic times as they are, persuading anyone that we need to spend a big pile of cash on hardware (and probably a couple more licences) is an uphill struggle. What I am really asking for is some extra vindication that what I have in mind is actually a goer, or if there is a better way of doing this.
My Problem
We have an 11 hour backup schedule, that has to cover about 100 separate systems (none with > 1GB NIC). The largest of which is a 7.2TB File Server (yes I know, throwing some more network at this is required) and the smallest a 4GB app box (we don't do consolidation !¬_¬). Overall we have to handle on average about 14TB of data in any one night.
The "new" server is a single processor server, with 2GB RAM and a 1GB NIC, with no on board storage, hooked up by 2 dual fibre channel cards to an ADIC Scalar i500 (which hosts 80 cassettes). This is all dedicated setup so its not shared with anything else. Now I know at this point you are probably thinking WTF! I know I did. There is a massive choke point at the network which is causing any number of missed schedules as well as some poorly provisioned backups that cause snapshot errors where the same vhost is having more than one client being backed up at the same time.
My Solution
Due to the massive network choke, I am proposing that we have a master server with two media servers. Each server would be laid out as follows
5GB NIC
8GHz CPU (prob 2GHz Quad's)
4GB+ RAM (Preferrably 8)
2TB Local Storage
Dual Channel FC (400MB/s min)
This would then have the Master server as a "streaming" system, with the 3 of our largest servers streaming through it direct to tape. It would also host the NOM as it is a streaming system, it shouldnt see too much of an overhead (or at least thats the theory) that would interfere with the streamed backups.
One media server would be a hybrid solution with it picking up the 3 next largest servers as a stream direct to tape, with a number of others running to the local disk storage.
The other media server would be a pure D2D solution during the overnight schedule.
During the day, when the tape drives are clear I can then run D2T backups from the media servers, and have the master server available for any restores that may be necessary.
Any comments and suggestions that would improve the setup are very welcome as this would be my first backup solution rollout.