Consumer disks trump enterprise platters in cloudy reliability study The chaps at Backblaze have cracked open a spreadsheet and fed it some log files again, this time coming up with the assertion that contrary to expectations hard disks designated as destined for “enterprise” use don't always outlast cheaper kit directed towards everyday punters. A recap for those who aren't familiar with …
With a better warranty and returns process as well. Also, you are going to struggle finding many consumer SAS drives if you want something that isn't going to flood crap onto your your control plane when it dies, taking your NAS / SAN node down with it (as the SATA protocol is designed to do).
"...potential failures can be identified before they become actual failures..."
Except you can't. Most failures occur without warning, compared to WITH some type of warning. The most reliable _statistical_ way of predicing failure, is knowing one drive out of particular batch has failed - statistically, the others in the batch will fail at about the same time. (perhaps not down to the minute, but over the service life of the drives, batch numbers are a reasonable indicator).
"...and better error handling/correction so they can get by longer in a degrading state to give you time to replace them..."
Enterprise drives are usually built more conservative in nature, which is where the reliability comes from. If it were a mere firmware change to improve error correction don't you think this would bleed down to domestic drives? After all, if you can offer a better MTBF than your competitor for the same price....
Enterprise SATA drives are functionally identical to consumer drives. I hope they get a bit more testing to get rid of the most egregious duds, but that's it. They don't have error correction, and the array is designed not to trust them. Unlike SAS, error correction on enterprise SATA is (or should be) done by the controller, for example by chucking in an ECC sector every few sectors, independently of any striping. That's why you get 1/9 less usable space with enterprise SATA.
Go check the numbers for nonrecoverable read errors per bit read on Seagate's consumer and enterprise drives.
The difference used to be a lot larger, probably for SATA they are getting close to be identical, but not quite. Back in the day you'd see about 5 orders of magnitude difference, instead of the one order of magnitude difference.
The formatting is different to accomplish this, more ECC bits, etc. And to the guy who said failures all happen without notice, you're an idiot. You've never once had a disk that started reporting errors but you were able to get the data off? If so, you're the most unlucky person in the world. There is generally notice before a drive fails. I guess you've never worked in an EMC environment, and have the EMC tech come along to replace a working disk, because their diagnostics show it is about to fail.
I was told by somebody in the trade to use 3.5" (or even 2.5") drives instead of 5.25".
The rationale was that even though you need fewer of the bigger drives, the overall reliability of the smaller drives was much greater because they were designed for laptops where they would be expected to receive rough handling - The larger drives were for desktop use where they would "not be subjected to abuse".
Apocryphal, but they claimed to have evidence based on warranty returns.
I've heard similar claims but the apocryphal claim for the improved reliability is that the edge of the platters isn't subjected to the higher stresses of the larger platter at the same rpm. While it sounds plausible, I'm skeptical until someone releases hard data and publishes their rigorous test protocol.
I have a couple of WD Raptors I bought and installed in striped raid the morning of my wedding in October 2002 (I had a couple of hours to kill!) that have been running almost constantly since that day. The have only been off when I updated my mother board (twice), during power failures, and when we moved into our house seven years ago.
No problems with them, ever.
Enterprise drives are not supposed to be more reliable.
They are supposed to be faster, typically spinning at 7200 rpm. In return, they sacrifice obviously more wear because of the higher speed. They are also supposed to be kept in a controlled and protected environment, and are typically deployed in RAID1/5/6 configurations where failure can be coped with.
Consumer drives on the other hand have to cope with rough handling, be it in tower cases under the desk which are regularly kicked or knocked, or laptops which suffer even worse. They are also deployed in environments where there is no redundancy and often no backup.
Of course consumer ones are going to be more reliable. Its an obvious consequence of the engineering brief.
Except the engineering brief is for enterprise drives to be more reliable because companies employ bean counters to watch expenses and consumers don't. Companies also have the large number of drives over which the bean counters can calculate those numbers.
Kudos to Backblaze for continuing to publish their data.
Your logic fails to account for the higher costs that enterprise drives are sold at, therefore allowing the beancounters to balance their equations of longer warranties, with greater claims against aforementioned.
The balance sheet of more claims against a longer warranty can easily, and often is in industry, rebalanced by simply charging more...which the drive manufacturers do indeed to for "enterprise"-level drives. If the Backblaze statistics hold true across the equivalent time sample, or conversely even if they do not, the costs for greater failures (engineering failure or time based wear) is simply amortized against the higher profits.
I would advise you to examine and learn the industry principle, best displayed by the "Ford Pinto memorandum", to see how beancounters justify this (common) accounting practice.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
