Disk drives: knowledge is power

There’s one thing that should frighten everybody who uses a desktop or laptop computer or server, and that is disk failure. Your storage should be the only part of the computer that you really care about. If not backed up, the information on your disk can be irreplaceable: when it’s gone, it’s very probably gone for good. Yet your precious information is entrusted to one of the few components in a computer that can break down completely, without warning, or literally wear out.

These days, there are two kinds of storage: traditional hard disk drives (HDDs), and so-called solid state disks (SSDs). HDDs contain highly delicate moving parts, whereas SSDs are related to the universally understood USB stick. The functional purpose is identical, but there the similarity ends.

It is important to understand these differences. HDDs are more capacious and cheaper, use more electricity, are heavier, slower, noisier, and more delicate than SSDs, but if treated gently, HDDs are generally understood to have a long lifespan. SSDs have a finite lifespan (measured not in years but in the cumulative amount of rewriting); you can flog them to death in months, but typical usage will result in many years of good service. A certain amount of anxiety attaches to SSDs, because of the known fact of their finite lifespan and uncertainty about what constitutes typical usage.

HDD and SSD reliability is shrouded in commercial confidence. However, in the last year, two studies have broken decades of silence. Backblaze, an independent company which operates more than 40,000 HDDs, has published statistics on which brands they find are most and least reliable, and the Tech Report website has spent months destructively testing a small fleet of different SSDs.

For HDDs, the lessons are as follows: 78% survive more than four years; the median is 6 years. HDDs have three distinct phases of failure: early failure of some drives when brand-new, then a minor amount of random failure in the ensuing years, and then finally they just wear out. The ‘bathtub curve’ model of failure was thus confirmed. Don’t let them get too hot, because that was found to increase the failure rate. HGST brand drives (formerly Hitachi but now part of Western Digital) are the most reliable by a clear margin. Western Digital’s reliability varies by model. Seagate’s reliability also varies by model, but is worst by a clear margin: the 1.5 Tb model had nearly a 14% annual failure rate.

The SSDs all performed flawlessly up to and, in some cases, far beyond the manufacturers’ claimed lifespans. However, there was a very disturbing twist. In almost every case the end came suddenly, with minimal warning, and made all the data utterly unrecoverable. In such circumstances, only a fool would operate an SSD beyond its rated lifespan or trust a second hand SSD. Maybe in the manner of their passing SSDs are not really very different to HDDs after all.

Now that you know all that, I hope you’re still a bit frightened, but much more empowered to make rational decisions on storage purchasing, backup and replacement.

As a postscript, there will be many gentle readers of this column who are scandalised by my spelling of ‘disk.’ My authority is none other than H. W. Fowler, who in the first edition of Modern English Usage described that as ‘the earlier and better spelling’ and declined further comment: as shall I.

David Spencer