seagate-"archive"-disks with raid6?
hi.
for a larger archive, we're thinking about buying six 8TB-Archive-Disks from Seagate to build a 4+2 Raid6-Array.
now we've seen, that these disks are told not-recommended for raid, because they lack ERC/TLER (like many desktop-disks)
unfortunately we didn't found any real actual information about behaviour of linux-raid with such disks.
so our first questions is, if anyone here already has disks (without ERC/TLER) with linux-raid running and can say something about?
other item is, - as we understand - ERC leads to a limited timeout on read-errors, so the wanted data of that read-access may recovered by the redundant data and marks that disk for soon replace (or a sync to an existant spare is started).
but this is also done, if that disk does not timeout and will be dropped by the controller, or is this wrong?
so the only benefit seems to be, that with ERC the disk is not dropped immediately, so errors on two further disks won't lead to a data-loss before sync (as long the additional errors are not corresponding raid-sectors)
but: how likely is it, that three disks have permamnent errors during the sync? (we already had two disks in the past, but never three at a time...)
how is your experience with multiple disk-errors?
regards
w.
Quoted 10 lines · wiebittewas
hi.
for a larger archive, we're thinking about buying six 8TB-Archive-Disks from Seagate to build a 4+2 Raid6-Array.
now we've seen, that these disks are told not-recommended for raid, because they lack ERC/TLER (like many desktop-disks)
unfortunately we didn't found any real actual information about behaviour of linux-raid with such disks.
so our first questions is, if anyone here already has disks (without ERC/TLER) with linux-raid running and can say something about?
I'm running two Seagate Barracuda disks in Raid-1. That, I believe is okay. I've never had any problem, but the disks are pretty new...
Quoted 5 lines
other item is, - as we understand - ERC leads to a limited timeout on read-errors, so the wanted data of that read-access may recovered by the redundant data and marks that disk for soon replace (or a sync to an existant spare is started).
but this is also done, if that disk does not timeout and will be dropped by the controller, or is this wrong?
This is right. 6 by 8TB is 48TB of disk. If the disk spec is standard, it says "expect one soft error per 10TB read", so a rebuild will get FIVE errors.
If you don't do anything special, each error will kick a disk out of the array ... OOPS !!!
Quoted 6 lines
so the only benefit seems to be, that with ERC the disk is not dropped immediately, so errors on two further disks won't lead to a data-loss before sync (as long the additional errors are not corresponding raid-sectors)
but: how likely is it, that three disks have permamnent errors during the sync? (we already had two disks in the past, but never three at a time...)
how is your experience with multiple disk-errors?
No experience here, but I'm planning to upgrade to raid 5 or 6, so I've been mugging up on it.
You need to increase the raid timeout, so the disk times out before the raid does. That's why you really need ERC, so you can make the disk time out before the raid does.
Cheers,
Wol
Quoted 3 lines · wiebittewas
for a larger archive, we're thinking about buying six 8TB-Archive-Disks from Seagate to build a 4+2 Raid6-Array.
now we've seen, that these disks are told not-recommended for raid, because they lack ERC/TLER (like many desktop-disks)
Not only that.
You might want to read:
http://www.spinics.net/lists/linux-ide/msg50641.html
These drives are different beasts than regular HDDs, and we seem to be seeing problems with them just the way SSDs were problematic in the beginning.
So you might want to reconsider using SMR drives for RAID use. I was considering them until I read up on them, and then I decided it was too early in the deployment cycle to use them for RAID use.
Signature · 1 line
Mikael Abrahamsson email: [email protected]
Quoted 14 lines · Mikael Abrahamsson
Quoted 5 lines · wiebittewas
for a larger archive, we're thinking about buying six 8TB-Archive-Disks from Seagate to build a 4+2 Raid6-Array.
now we've seen, that these disks are told not-recommended for raid, because they lack ERC/TLER (like many desktop-disks)
Not only that.
You might want to read:
Quoted 8 lines
These drives are different beasts than regular HDDs, and we seem to be seeing problems with them just the way SSDs were problematic in the beginning.
So you might want to reconsider using SMR drives for RAID use. I was considering them until I read up on them, and then I decided it was too early in the deployment cycle to use them for RAID use.
I didn't know these were shingled drives ... does the OP know what a shingled drive is?
Basically, to get decent performance out of these drives, you have to stream data at them - they're more like a tape-drive than a random-access disk. So. They're great for backups, much less so for normal use.
Cheers,
Wol
Am 13.06.2015 22:11 schrieb "Mikael Abrahamsson" bzgl. "Re: seagate-"archive"-disks with raid6?":
Quoted 1 line
thanks for this link, which shows us, that there might be better solutions in near future.
Quoted 1 line
These drives are different beasts than regular HDDs, and we seem to be seeing problems with them just the way SSDs were problematic in the beginning.
if "different" means that they use SMR - yes, we know this. Our current use is mostly write-once/read-many, so the described additional timeouts might not happen here. nevertheless, the cooperation beetween filesystem and disk, which is mentioned at the given link, sounds better, so we'll try to wait some weeks or might find a temporary solution. (perhaps this new series will have ERC/TLER enabled, when they let the kernel do the SMR-control...)
regards
w.