Western Digital Recognises SMR is an Issue for NAS Drives
For most of us, we have become accustomed to having our files available to us at the touch of a button without much care for the future. We expect and demand our data to live online, for free, in perpetuity. While irrational, it does mean that our ‘lives’ are lived online, which in turn means we are a digital culture. With this new transcultural landscape backing up data becomes ever more critical. It means more than just having a spare 8Tb drive to save some home videos, but to know that it will be around for another 5 years, at the very least, so you can then transfer it to another medium – hopefully something better.
To meet this change in data habits, Western Digital, maker of the world’s first hard drive, pushed limits of data density by introducing shingled data tracks (SMR). A process where tracks of data are layered so that they slightly overlap, giving the spinning hard drive even more density. The only problem is that they seem to work less efficiently in large data arrays, tend to be slower, and were not labelled on consumer, gaming, and corporate hard drive product lines, which might be a bit of a problem.
Shingled vs discrete tracks
In traditional systems spinning hard drives are tasked with writing 1s and 0s along tiny tracks, which is very much like the tracks on a vinyl LP. These long, thin tracks keep your data recorded and allow for a complete read the next time requested. Shingled Magnetic Recording (SMR) technology was developed to pack even more storage in the same space to allow for increased capacity – some 25% more. This results in larger drives that are also cheaper. The typical tradeoff is that these drives are slower for both reading and writing, and in some cases, less reliable.
Each time a data block needs to be rewritten an entire zone requires rewriting, for the equivalent in standard drives no such step is needed. Also, SMR drives have space reserved internally that acts as a scratchpad for the drive to temporarily store data and then later makes writes to the drive-in shingled format.
We can see how this might be an issue as smaller files are easily written to the temporary drive area and then transferred, but what if you are working a NAS, RAID configuration, or ZFS vDev, or all three, which might have a continuous IO of terabytes? This is where the technology has appeared to become a bottleneck and how WD (and now Toshiba) have failed to label their SMR drives. Alan Brown from UCL Mullard Space Science Laboratory, the UK’s largest university-based space research group, told Block and Files
In the case of ZFS, resilvering isn’t a block-level “end to end” scan/refill, but jumps all over the drive as every file’s parity is rebuilt. This seems to trigger a further problem on the WD40EFAXs where a query to check a sector that hasn’t been written to yet causes the drive to internally log a “Sector ID not found (IDNF)” error and throws a hard IO error from the interface to the host system.
Alan Brown, UCL
In practical terms, it means that drive IO would be inconsistent as data is being continuously shuffled between the safe conventional holding area to the final SMR destination. Normally, drive speeds for resilvering, a term for ZFS filesystem rebuilding, would operate at near maximum drive speed, appears to only be half that – if lucky. Naturally, this extra wear will mean that drives are less robust as previous iterations and make one wonder if the trade-off was really worth the choice.
On WD’s corporate blog they have admitted to seeling SMR drives without the appropriate labeling which has only caused further headaches for data hoarders. Knowing the exact internal topology and characteristics of a drive is incredibly useful when creating long-lasting RAID arrays. For instance, when I build my own arrays I tend to use drives that use a single platter for I know that the internal layout would be simpler, have fewer moving parts, and if the need arises, I can perform a hard drive platter transplant to save the data. Thankfully, my arrays have never needed this.
In any case, the outlook for HHD is still bright as the road map for the now 60 year old tech is bright, with 50Tb drives forecasted for 2022.