Your question regarding best practices is harder to answer than you might first think because best practices for me won't be best practices for you. (Different business risk regarding data loss, among other things.) If your site is subject to regulatory issues (as mine is), you find that you cannot allow data to NOT be backed up.
The decision to do backups at all is a business risk question and for any organization with really valuable data, it is hard as hell to justify doing no backups. It is almost a "given" that you will run a backup. So the next question is, "how far back must you go?" and that is another business risk decision. How much does it cost you to go back x number of days? When you factor in the cost of keeping those media for x+1 days (minimum), there comes a breakpoint where you say "we can afford to track up to x days before it is cheaper to just type it in again" (or do without it).
The problem of eliminating tapes (as noted by another responder) is that if you WERE using tapes, you just have to replace them with something else. We have tried many other solutions - for example, a WORM-drive CD Jukebox (3rd party vendor) - but we were not satisfied with reliability vs. cost vs. ease of handling. The cost of having dismountable/removeable volumes leads to questions of why you have such things and aren't using them for live data? Sounds crazy, right? I can't tell you how often in the real world you run into pointy-haired managers a la "Dilbert" who can understand backup to tape but can't justify backup to dismountable disk.
There is a trade-off in terms of how many volumes of backup media you can keep so that you can do things like - go back in time to pick up files lost several days ago but the loss wasn't discovered until today. I wish that never happened but sadly, it really does happen. I've had to go back in time a couple of weeks in the worst-case scenarios. Yes, I have clumsy users. But isn't that expected?
For us, the decision is resolved by staggering our backup pattern using a tape-based storage system.
1. We back up incrementals daily
2. We do a full backup weekly
3. Once per month, we set aside a weekly full backup to become the longer-term monthly backup.
4. At the end of a year, the monthly tape goes back into rotation.
5. Once per month we evaluate what is actually on the tape (i.e. verify that data can been identified)
6. Once per month we recover files at random to verify that they are indeed recoverable. Using the full backup as a source, we can recover a file not changed recently and compare the backup copy to the still-resident copy. Yes, it is an issue in sampling theory as to how many files to check.
As to using big block sizes, I concur with the comments about streaming media being best used if you can build big blocks. Not to mention multi-buffer counts being as high as you have memory to support. The thing that eats your lunch speed-wise is tape rocking. I.e. your streaming medium ran out of data so you have to stop the tape, rewind it, get another buffer ready, start the tape looking for the end of the previously written block, and switch from read to write mode. That eats your tapes, your time, and your data reliability badly.
Security+ Certified; HP OpenVMS CSA (v8)
