Discussion in 'Article Discussion' started by Gareth Halfacree, 13 Jan 2016.
Beats WD to the punch.
Importantly, this differs from some other 8GB drives available to consumers in two ways:
- It's a regular PMR drive, not SMR, so does not have the strange write characteristics of Shingled Recording.
- It has an URE rate of 1/10E15 not the more common 1/10E14. This is vital due to the size of the disc being over 6E13 bits.
And would you please explain what that all means?
SMR = scary weird magic based in no small part on how SSDs work, which lets the drive store data over existing data and means things get written in a really, really strange way. It allows for cheap and massive drives, but by crikey it's odd - so much so, in fact, that Seagate offers an entire filesystem (actually a hacked-around copy of ext4) specifically for use with its SMR drives, to work around some of the weirdness. How weird is it? Well, I spent more on a 6TB drive than I could have bought the Seagate 8TB Archive SMR drive for specifically because the technology wigs me the heck out.
The error rate stuff is the number of bits you can read before you're likely to encounter an error. When this number is bigger than the number of bits on the drive, you're golden: you can read the whole drive without error. When the number is smaller than the number of bits on the drive, you've got issues: the manufacturer is basically telling you that there's a good chance that you'll get an uncorrectable read error should you try to read the contents of the disk back in its entirety.
Due to physical limitations, a hard drive's write head is bigger than its read head. Traditionally, this means the width of a magnetic 'track' of data in a drive is the width of the write head.
With SMR, the tracks are made the width of the read head so that they 'overlap' (from the point of view of the write head). This means that if you want to change data on one track, your write head will accidentally write over the other overlapping tracks (because it is too big), so every overlapping track in that area has to be read and re-written. There are gaps in the shingling so that you don't have to rewrite basically the entire drive every time.
The diagrams on http://www.storagereview.com/what_is_shingled_magnetic_recording_smr/ make it easier to understand too. Obviously this is a simplification, but it's clearly not black magic.
As you can imagine, this makes writes slow. Like SSDs, there are techniques to making this faster which are dependent on making the operating system aware of what is going on underneath. This is a bit like how you have TRIM commands for SSDs. But compared to SSDs I think the implications of your OS not being aware of the SMR drive are not as serious. I'm not so sure about that though.
The OS not being aware of an SSD is not particularly serious. You see some reduction in max IOPS due to a suboptimal sector size (any file that is between 512b and 4kb in size may be affected by this, but smaller or larger will not), and may see some extra Write Amplification depending on how well the controller handles sub-4kb writes. an SSD that utilises copy-on-write exclusively (and has a nice chunk of spare area available to do so) will see little impact from lack of OS optimisation.
SMR on the other hand will see MASSIVE issues if the OS is not SMR aware. Mainly this is because random write latencies can be several seconds due to the need to re-write a whole swathe of sectors at a time, every time.
Separate names with a comma.