Sooo – now list the manufactures that had high failures and complete the information delivery for the article. This would be important since the SMART parameters themselves have been determined (by Google) to be useless!

Good never mentions that drive manufacturers. Their article also points out that in general failure rates follow drive batches. But tracking a batch and testing can be very difficult (if not impossible).

I also wouldn\’t say that google determined the SMART parameters are useless. Rather the SMART parameters by themselves cannot be reliably used to predict _all_ drive failures. They can capture some some drive failures but you need to use some heuristics to use the parameters for making predictions. Even if you use all of the heuristics and other metrics that Google used, they still couldn\’t predict all of the drive failures (or close to them).

Please read the Google paper for more details. It\’s a very very interesting and useful paper.

Jeff

The argument on the drive manufacturer side is that if that drive was to be resold, the potential buyer should have some warning that a problem is imminent. But if we applied the same logic to cars, there wouldn\’t be enough room in all the landfills on the planet to dispose of all the unsellable cars flagged as \”failed\” just because they have some minor mechanical issue.

Would that section title be from \”Brain and brain, what is brain\” ST:TOS \”Spock\’s Brain\”

Far too simplistic. It may be one particular model that\’s a lemon. It\’s probable that this didn\’t become apparent until after the manufacturer had stopped selling it (i.e., it\’s obsolescent). Or it may be a manufacturer who has fallen foul of a batch of dud components that could have been sold to anyone. Many other possibles.

The fact is, every new drive on the market is in many senses a prototype. By the time it\’s been in service for long enough to prove reliable, it\’s also obsolescent. The manufacturers do accellerated ageing tests, but these are a poor approximation to the passage of years of real time.

The moral – if you construct RAID arrays out of disks with near-consecutive serial numbers, you maximise your chances of a multi-drive failure! To eliminate several common-mode failures, build an array out of (say) one WD, one Seagate and one Hitachi disk. For a bigger array if you run out of manufacturers you can use different sizes or series from the same manufacturer. It may even reduce the risk to buy the same drive from two different suppliers – the serial numbers are less likely to be near-consecutive, you may dodge a batch problem that way.

Other morals – if a drive in a RAID array fails assume it\’s a pointer to another being about to go the same way! And with drives as cheap as they are, consider RAID-6 rather than RAID-5, just in case.

Ding, ding, ding!!! We have a winner. I love the line from the Spock\’s Brain episode where they are questioning the woman and she yells,

\”Brain and Brain, what is Brain?!!\”

I just think it\’s such a great line to use here and there.

Thanks!

Jeff

The brain has long been mysterious in this way. It was purchased
from an outside vendor. It fits neatly into a slot at the top of
the case. It picks up signals from the rest of the system and
provides output signals that are used to navigate and maneuver the
entire device. In a classic Star Trek\’\’ episode, a nasty alien
stole Spock\’s brain and plugged it into a central control station so that it would run the air conditioning or something. Complex plot
twists led to the immortal line: Brain, brain … what is brain?\’\’