Best backup plan - multiple generations of backups?

[dsiomtw]

I've run RAID 1 forever and have always had a simple backup plan using Acronis where I'd just clone my OS drive to a NAS device once a month and my data drive (only about a gig) would get cloned every night.

Recently I switched to SSD drives, and after seeing how big of an improvement the fastest drives made, I really want to test out a RAID 0 and go even faster.

Question is, what is the best backup plan when running a RAID 0?

If I stuck with my current setup, I would be completely hosed if a drive or controller died while a backup was in progress.

Never really had a reason to think about it much, but now that I do I'm shocked that Acronis doesn't let you keep X days worth of backups and automatically name them backup1, backup2, etc. Backup "generations" I think this is called.

Am I totally missing something here, or am I supposed to cobble together my own custom scripts to handle backup renaming/archiving/rotation/deletion/etc???

 

[sminlal]

Don't expect that much of a real-world performance improvement with RAID 0. Yes, you can get significantly faster transfer rates, and those will show up nicely on benchmarks. But in the real world, most of the speed-up you see in booting and starting applications with an SSD comes from their much faster access times, not their transfer rates. And RAID does nothing to improve access times.

 

[dsiomtw]

Hmm, not sure I can agree with that. With my Intel 320 I get about 275 sequential reads. In a RAID 1 I get 450-500 and I can definitely notice the difference. In a RAID 0 it should be even faster.

 

[sminlal]

Sequential benchmarks measure transfer rates, and as I said they are indeed faster with RAID. That might help you if you spend a lot of time copying, reading or writing large files - but that's not a very typical workload for an SSD. For booting and starting applications you get the most benefit from faster access times, and RAID won't help with those.

If all you're looking for is better benchmark numbers, then by all means go ahead. But if you're looking for real-world performance improvements I think it might be prudent to lower your expectations.