==> In article <[EMAIL PROTECTED]>, Alexander Lazarevich <[EMAIL PROTECTED]> writes:
> Does any TSM'er use disk based backup? We've considered it, and are experimenting. Here's what to keep you eyes out for: - Sustained outgoing bandwidth - Sustained outgoing bandwidth in multiple streams - Sustained outgoing bandwidth when other stuff is writing So far, my observation is that, if I can just line the tapes up correctly, restoring filesystems from tape is as fast or faster than disk. This distinction gets thrown into -sharp- relief if you have to do more than one restore at a time. If you've got (say) 85MB/s bandwidth out of your disk plant that may seem nice, but if aggregate bandwidth for three restores is still only 85, that might quickly become a problem. Even more importantly, if you're restoring and someone backs something up, in some architectures this totally blows out your cache, and you get big bandwidth hits. Not to sound too mantra-like, consider the restore case first: that's when you're really going to care about it. Assemble your 85% confidence-interval worst case restore scenario, and subject your loaner infrastructure to it. - Allen S. Rout
