Is it even possible to make solid reliability conclusions?!?

[ZapNZs]

It seems that for every 1,000 articles there are about the performance of drives, there is less than one about long-term reliability. That's causing me some frustration because I am interested in reliability and almost apathetic to performance.

For a very long time, I used 1tb WD RE4 drives for both single & redundant disk backup solutions. I've configured over 100 for clients, and the drives have performed wonderfully. However, when they were shifting over from the RE4 to Re, there were availability issues with the smaller sizes. Needing two drives fast in the 1 TB flavor, I went for the Seagate Constellation. Well, three years in and both are having some issues. Since RE4 stocks are dried up, I need to shift to something else.

Reliability testing with drives seem to be very difficult. Large sample sizes are needed to draw conclusions, details on the environment a drive is used in may be incomplete, there must be data collection/entry methods to catch reporting errors, tracking of which specific model is used (given a Maker could produce several variations of the same model/size, but with variable features such as more aggressive power management that could change wear outcomes), an ability to track changes in fabrication, and the ability to determine when other external factors may influence outcome (for example, a Tsunami destroying a fabrication or materials hub.) Heck, I found that the 1 tb RE4 was MUCH more reliable than larger sized RE4s, and if the test results from the same drive model in two different sizes have major variations, that's a major kick in the generalizability gems. Even if larger organizations have huge quantities of drives, and are willing to swallow the potentially massive costs for in-house research, I imagine most don't publish that information (especially if the results indicate that a specific brand/family of drives are really, really bad.) The Google study is great, except they aren't providing certain info needed to draw conclusions on specific models.

I'm currently debating between the HGST UltraStar, WD Re, or possibly one of the other enterprise-class WDs. With WD, their entire enterprise line was recently refreshed and so I am finding so, so, so little data on it. With the UltraStar, it seems to be highly regarded but not frequently tested.

So my question is whether or not you feel there is enough evidence to draw solid conclusions in this area? And how solid?...like a "this drive is good, but this drive is better" or "these types of drives are usually reliable"? Thx!

[LarrySabo]

Have you read Ross Lazarus's February 2016 "Survival analysis of hard disk drive failure data" report on his analysis of BackBlaze's data?