Epic Hard Drive Failure

[peterb]

I always thought that decreased performance and could ultimately lead to extra stress on some/all of the drives in the array due to different physical drive characteristics and spindle syncing?

I wouldn't have thought so. The drives aren't in physical sync, only logical sync. There is no guarantee that data written to sector nnn on one drive is written to the same sector on another. And with both OS and on-drive cache acting as a buffer, I can't see it being too much of a problem anyway.

But the advice to use drives from different batches is certainly valid.

i hope Samsung drives are not really really bad atm... i have 8 of them running in various flavors around the flat!

All RAID'd though so it wouldnt be too bad, dread to think how long the RAID5 with 4 x 2tb drives would take to rebuild though :/

I don't think Samsung drives are bad. They must sell hundreds of thousands of drives - with that number there are bound to be the occasional failure.

I would guess it depends on the mechanism (hardware controller etc) and configuration. For software raid (mdadm) the priority of the rebuild is configurable - but a 1.4Tb RAID 1 takes about 18 hours or so on a 1.8Ghz processor with the rebuild running at low priority.

Why don't you test it and see? (After taking a current back-up of course ) It is always useful to test these mechanisms under controlled conditions so that you can have confidence in them if they ever are required 'for real'.

[Biscuit]

I wouldn't have thought so. The drives aren't in physical sync, only logical sync. There is no guarantee that data written to sector nnn on one drive is written to the same sector on another. And with both OS and on-drive cache acting as a buffer, I can't see it being too much of a problem anyway.

But the advice to use drives from different batches is certainly valid.



I don't think Samsung drives are bad. They must sell hundreds of thousands of drives - with that number there are bound to be the occasional failure.

I would guess it depends on the mechanism (hardware controller etc) and configuration. For software raid (mdadm) the priority of the rebuild is configurable - but a 1.4Tb RAID 1 takes about 18 hours or so on a 1.8Ghz processor with the rebuild running at low priority.

Why don't you test it and see? (After taking a current back-up of course ) It is always useful to test these mechanisms under controlled conditions so that you can have confidence in them if they ever are required 'for real'.

Well its running on a PERC 5/i. Testing it isnt really an option at the moment as we are talking about ~3 TB of stuff. Was bad enough transferring data across form about 6 different hard drives just to set it up. Your totally right thoughi should get it backed uip and tested