If you want to learn which hard drives to buy, don’t look at other consumers. Ask a firm which spins a lot of them all the time, like a cloud service provider.
Thankfully some organisations which do keep lots of hard drives spinning are kind enough to tell us about the performance of their drives: Backblaze for one, which reports every quarter on its experience. Its latest report gives figures for the period January to March 2016.
Over that period, Backblaze used over 61,000 hard drives to store encrypted data for their customers. Overall, these drives were in use for a total of more than five and a quarter million days, giving an average of 58,000 drives in use at any time. These are normally set in storage pods, each containing 45 hard drives, which are then assembled into a vault totalling 1,200 drives. As a consequence, Backblaze tends to purchase drives in batches of five to ten thousand at a time.
The total number of drives which failed was only 266, or slightly less than three drives each day, or an annual failure rate of 1.8%.
Backblaze’s drives are likely to be much harder worked than yours, but equally their data centre will have much better controlled environmental conditions (they probably use more energy to cool the centre than to run it), and their drives will very seldom be powered down. Other studies of large numbers of drives in data centres show the importance of keeping them at a steady, cool temperature, and that repeated power on, power off cycles are likely to shorten working life dramatically.
Backblaze used drives manufactured by four main suppliers: Hitachi (HGST), Seagate, Toshiba, and Western Digital (WDC). Their capacities ranged from a few 1.5 TB units up to 8 TB, with the most popular being 2 to 4 TB. Failure rates varied widely across different models from the same manufacturer: for example, nearly 2,000 Seagate 6 TB drives suffered not a single failure in the quarter, but Seagate ST4000DM000 4 TB drives had a failure rate of 2.5%.
Other studies have shown how such wide variation is not confined just to different models from the same manufacturer, but batches of the same model and capacity vary widely too. The tendency for drives of one batch to fail at about the same time is the major reason that I recommend that, when buying multiple drives, say for a RAID array, you buy them from different batches. Otherwise you could find that all four drives in an array fail within a few weeks of one another – a very scary situation.
My own personal preference, based on experience over the past decades, has been for Hitachi and Seagate drives, and I try to avoid Western Digital. I do not have such objective evidence as that in Backblaze’s figures, which show that Hitachi models have generally lower failure rates than those of the other manufacturers. The highest annual failure rate of any Hitachi drive was in the HDS722020ALA330 2 TB model, with a rate of 1.7%, slightly below the overall average of 1.8%.
Backblaze’s cumulative figures are fascinating, but much more difficult to interpret, because of the change in their use of drives from different manufacturers, and of different capacities. Over their first year, they had hardly any drives of greater than 4 TB capacity, and those were almost all made by Hitachi or Seagate. Models which stand out include:
- Seagate ST31500341AS and 541AS 1.5 TB have had high failure rates throughout, and have largely been wasted out;
- Most Hitachi models have maintained low failure rates throughout, and have lasted the period without failure. The only exception were early HMS5C4040BLE640 4 TB models, but they are now performing much better, in the latest quarter having no failures among over 3,000 drives;
- larger capacity drives do not have higher failure rates. If anything, their annual failure rates are lower than those smaller than 4 TB.
One of Backblaze’s best-performing HGST HMS5C4040BLE640 4 TB drives would cost you anything from £122 to £250 depending on supplier.
Currently, Backblaze mainly buys 4 TB drives from Hitachi and Seagate, because they are most readily available (unlike Toshiba and Western Digital models), and most cost-effective in terms of capacity.
It is also worth noting that Hitachi Global Storage Technologies (HGST) acquired IBM’s hard drive business in 2003, and turned the Deskstar design from disaster (‘the Deathstar’) to great success. In 2012, HGST was bought by Western Digital, and has continued to develop and sell its own highly successful brands.
Thanks to Andy Klein and to Backblaze for sharing these data. They are invaluable.