After my girlfriend‘s Powerbook crashed, taking with it several months of her un-backed-up data, I decided enough was enough with my own antiquated backup hardware (ExaByte 8505 8mm tape drive, 5GB/10GB) and I bought a DLT7000 (35GB/70GB) drive off eBay, thus increasing my backup capability by seven times. With eight tapes, the whole adventure cost me approximately CAD$150.
I started thinking about what the purpose of hardware compression on tape drives is. In principle, it seems like it’s a good idea; offload compression, which is a CPU-intensive activity, onto the drive. The only problem is that it makes the estimation of whether or not a backup will fit onto a tape a virtual impossibility. I want to know, before I even start writing to the tape, whether or not a backup is going to fit. I don’t want to start writing to tape and then, 2 hours later, find I just hit End Of Media. It’s not something that you can recover from.
I don’t see a technical solution around this problem, so what I do is turn off hardware compression and just gzip the data to a holding disk. This is one of the great features of AMANDA; you can stage the entire backup to a temporary disk, and then write the backup to tape from that disk.
So, as far as I can tell, hardware compression is not very useful; it seems like a scenario where solving one technical problem (moving slow compression activities onto hardware) creates another (inability to know a priori if you’re going to run out of tape before you start writing the backup).