I am using Western Digital AV-GP will i have problems.

[ethereal]

Recently i was advised that these drives were endangering my data - is this true ?

 

[jgreco]

Their characteristics are less well-defined than the commonly used drives.

The AV-GP and other surveillance/DVR targeted drives tend to prioritize responsiveness. It is important in a DVR application that the device continue to accept streaming data at a high rate of speed and not pause for a significant amount of time when a write error is encountered, and likewise for playback, dropping a sector of data is not a significant crisis if all it means is a momentary glitch in a video frame. To that end, those types of drives typically do not have the emphasis on error detection, retry, and correction that storage array drives do. It is unclear on what exactly is presented upstream to the host platform in the form of errors, and this sort of ambiguity is not usually the sort of thing we encourage in a disk drive for ZFS. They have other characteristics that are reminiscent of WD Green drives, so you also gain all those issues such as needing to make sure you use WDIDLE set to disabled.

 

[ethereal]

I am using the drives in RaidZ2. So it has two copies of the data. If there are sectors of data that is incorrect then this should show up during my weekly scrub. I have had these drives between 1.6 years and 0.6 years - with weekly scrubs and long smart tests. There have been no repairs and no known data errors - there are also no bad sectors viewing the smart data

 

[jgreco]

Right, that's what we'd hope to be the case. I'm not concerned about how it works when everything is working as it should. If they screwed that up, well, fsck them. I'd be more worried about how things work when something goes wrong, since that's where a lot of the differentiation in behaviour is between the models.

 

[ethereal]

okay - let me know if i understand you now.

i thought you meant the error correction was so bad that the drives would write data that was inconsistent and i would find it during a scrub.

do you mean i will only have problems if the drives have a bad sector?

 

[jgreco]

No. I mean the error correction strategy on those drives is different. Look, I totally expect that under normal operation you wouldn't be able to tell them from a NAS class drive, and in that regard they're fine.

When you eventually develop a bad sector, though, then the questions pop up.

Writing a bad sector.... on a typical drive this will force some retries and then a reallocation of a spare sector if it cannot write the block to the defective sector. What happens in that case with these drives?

Reading a bad sector.... on a typical drive, the drive will attempt retries in order to recover the data. In a nonredundant system like a desktop PC, the disks try really hard to recover the data, maybe even locking out other operations for ten seconds or more. We know that's not ideal for ZFS which is why I like drives that support TLER. But the surveillance drives go the opposite direction, rapidly abandoning any attempt to recover from errors.

And the meta issue is, how will ZFS respond to whatever the drive reports back when an error occurs? Are the error responses the same? Are they maybe suppressed in some way in order to make a DVR appliance happy? Do Seagate DVR drives respond the same way as Western Digital DVR drives? Honestly no one here is really tracking this sort of thing. The big thing I'd be afraid of is that some unusual response from the drive causes ZFS to freak and drop the drive out of the array over what is essentially a trivial misinterpretation of status.