Storage – either solid-state (SSD) or hard disk – is usually one of the more expensive purchases after your Mac. If you have an external RAID unit, replacing its drives can cost as much as your Mac. Making a good decision will not only save you money, but will safeguard the data which you trust to that storage.
Fusion Drive, SSD, or hard disk?
The first choice is whether to opt for faster and significantly more expensive SSDs, or more capacious, cheaper, and slower hard drives. Here, speed and cost are going to be the most important factors. Combining the two in a Fusion Drive seems to have worked very successfully for Apple, and you can, with a bit of careful work at the command line, create your own, using macOS CoreStorage.
Not all SSDs are made equal, though: some users of Apple’s Fusion Drives and of plain SSDs have suffered premature failures. At the moment, it is hard to know how or why these occur, or even whether they are hardware failure or artefacts in SMART indicators. All SSDs have a fixed number of times that each storage unit can be written to, but in practice that should far exceed the expected life of a hard disk. That holds good so long as wear is levelled across all the parts of an SSD, something which should be performed automatically.
Fusion Drives are also not normally optimal for storage of backups, such as those made by Time Machine, but better suited for use as your startup volume.
Filling a RAID with high-capacity SSDs is an extremely expensive proposition. If you need to stream very large amounts of data, such as high-definition video, they may be your best option.
APFS, Apple’s new file system being introduced with High Sierra, is another factor worth bearing in mind. APFS has been optimised for use with SSDs, so you should see significant performance improvements when using APFS rather than HFS+ on an SSD. Whether there will be any significant improvement on hard disks remains to be seen.
Hard disk manufacturer
Datacentres and others who use very large numbers of hard disks have long been aware that their failure rates vary widely between different manufacturers, capacities, models, and even batches of the exact same model. Fortunately some have made their experience available openly, and the best information is that provided each quarter by Andy Klein of Backblaze.
I have previously looked in detail at Backblaze’s usage and failure statistics, and here update my conclusions from their latest figures, for the last four years.
Backblaze has used large numbers of drives manufactured by three different companies: HGST, which is now owned by Western Digital, Western Digital itself, and Seagate. Experience consistently shows lower failure rates for HGST drives relative to Western Digital and Seagate. This is not true for all HGST models, though, and some Seagate models have been performing rather better than others.
It is strange that, now HGST is owned by Western Digital, HGST drives still seem to have lower failure rates than those of its parent company. However, HGST drives originated from IBM’s catastrophic Deskstar design, which failed so early that they became known as Deathstars. Their underlying design was novel, but it took Hitachi to iron out the problems and turn them into very reliable units.
So if you want to buy by brand, you should be better off looking for HGST.
Over the years, Backblaze has steadily increased the capacity of the disks which it uses for storage, and currently specifies 8 TB. One fear is that such large drives might be less reliable than more modest capacities such as 3 or 4 TB, but there is no evidence that that is the case. HGST 4 TB models have clocked up remarkably low failure rates over many millions of days use, and Backblaze’s experience with larger drives is more limited at present.
If you want to make the safest choice, you would probably do best with an HGST 4 TB drive, but if you could do with the additional storage, don’t be scared to go up to 8 TB.
Enterprise or regular quality?
Various ‘quality’ designations are now used for hard drives, but there is only one which should be of particular significance: if you’re going to install a hard disk in a laptop, ensure that it is designed for more frequent spin-up and spin-down cycles, and ideally has better resistance to mechanical shock.
For desktop use, what is more important is the period of warranty. The great majority of models are now only offered with three year warranties, and you should aim to replace them when they are little more than three years old, unless they have been little used. Product engineering is closely tied into warranty duration, and a lot of products are engineered so that their risk of failure is very low until the warranty runs out. Thereafter the risk rises, and failure should be expected. Few hard disks are now likely to carry on working faultlessly for longer than five years.
There’s a problem peculiar to RAID storage which needs to be taken into account. Disk manufacturing achieves great consistency in quality within each batch, but quality varies more between batches. The practical consequence is that disks made in the same batch tend to fail for the same reasons at roughly the same time. This is enhanced by the fact that disks operated together tend to experience the same environmental and load conditions too.
Most of the more popular RAID levels are designed to cope with single-disk failure, but if two or more disks fail at about the same time, data may be lost. One good plan to minimise the chances of multiple disk failure is to select hard drives from different batches (but identical models and specification). This can be fiddly to arrange, but is well worth the effort.
Hard drive care
Once you get your storage, there are some basic steps which you can take to ensure that it does survive past its warranty period, if not into graceful retirement soon after. Avoid all mechanical shocks, particularly when the drive is spinning, and ensure that it remains at an even temperature towards the lower end of its operating range. Overheating drives is a technique used to age them rapidly, and substantial temperature changes also tend to increase wear.
Datacentres inevitably keep their drives spinning at all times, until they are replaced. Their annualised failure rates are usually much lower than those experienced by normal users, and one reason for this is that most users let their systems put hard disks to sleep, so spinning them down. Hard disks suffer their greatest stresses when being spun up and down, and those stresses are reflected in increased likelihood of disk failure.
If you want a desktop hard disk to last as long as possible, keep it spinning for as long as you can: don’t let Energy Saver put it to sleep, and don’t shut your Mac down at the end of the day.