No matter how much storage you have, given enough time, you’ll always need more. As part of my new build, I decided to follow my standard procedure of upgrading part of my storage to larger, new drives for increased speed, capacity and reliability while pushing older, smaller drives out to cold-storage such as to save energy. For the most part, the strategy works out, but as the drives get larger, they typically get more expensive per gigabyte. This means the outlay increases every time as storage prices have seemingly stagnated.
In the past, I had already reviewed the Seagate Archive 8Tb series of drives in 2015, which proved to be an interesting drive because it combined large capacity with low cost. The tradeoff was inconsistent I/O performance borne from the need to utilize SMR techniques to “cram” more data into the same platter area and preclude the need to use expensive helium drive manufacturing techniques.
Fast forward to the tail end of 2017, and facing the same dilemma of getting large bulk storage at the lowest possible price, I scoured the local computer shops for solutions. As it turns out, the Seagate Archive 8Tb drive was still available but relatively unpopular at around AU$370 (a higher price than when I bought them in 2015). The other low-cost option was the Seagate Skyhawk Surveillance 8Tb drive, at around AU$360. Being AU$10 cheaper per unit, I couldn’t see any immediate downsides, and so I opted to obtain two units.
These are positioned as a surveillance optimized drive, with “ImagePerfect” firmware optimized for the application, which presumably indicates prioritizing writes over reads, ensuring compatibility with DVRs/NVRs, maybe support for streaming write/read commands and perhaps a tweak to error-handling behaviour (as video can be somewhat tolerant to data loss with broken playback as a “glitch”, and rapid drive recovery would prevent data loss from queued writes or the drive being considered “offline”). The drives offer multi-bay support with rotational vibration sensors, a 3-year warranty under 24/7 power-on-hour conditions and 180TB/year workload. Claimed maximum non-recoverable read-errors per bits read sits at 1 in 10^15, which is a “typical” figure, although some of their smaller drives still claim 1 in 10^14.
Of course, I wasn’t intending to use them in surveillance applications. After all, I’m very much against artificial market segmentation through software differentiation and I’m convinced that under “general” workloads (i.e. non RAID, low utilization factor, at home), absolutely any drive should work just fine. They are just a block storage device after all – store data in block X, read data from block Y.
As with most newer Seagate drives, they come in a static shielding bag which has seals along the long side of the drive. This is my first drive with their new “Guardian” branding – e.g. Barracuda Pro, Firecuda, Ironwolf, Skyhawk.
There is a seam at the back as well. The packaging really reminds me of a chip packet.
The drive is very solid and weighty in the hand, but if you’ve held the Seagate Archive series drive, you’d be mistaken for thinking they were the same. The drive weighs a whopping 650g, and basically makes use of almost all the space available to it, with a rear casing that’s almost flush to the bottom screw holes. The label on the top features their new logo, but also has QR codes for convenience in checking the drive’s information. The drive was Made in Thailand, with model number ST8000VX0022 and a firmware revision of AV01.
Identical to the Seagate Archive, it misses out on the centre screw holes on both sides. If your case requires them, then this drive might not be the one for you.
At the rear, we have the standard SATA and power connector positions along with a debug port. At the front, we have a serial number label for quick identification, and a stepped region of the case (which might be used during automated manufacturing).
As the ever-curious sort of person that I am, I decided to partially take apart the drive (non-destructively) to see what was on the PCB.
Just like the Archive, the PCB has this L-shape with two thermal pads transferring heat from the controller and motor driver. A foam insulation sheet with translucent plastic covers the board to prevent shorting to the case – there is a clearance where a large bunch of MLCC capacitors are positioned near the power handling section of the board.
Removing all of these adornments reveals the main board which is virtually identical to that of the Archive drive with the exception that an EEPROM has been replaced with a surface mount device instead. Three vibration sensors can be seen (two large, one small at orthogonal axis). Next to the main controller is a piece of Samsung DRAM for cache.
With an addition of a smear of thermal grease, the codes for the chips can be clearly resolved. As it turns out, they’re exactly the same as on the Seagate Archive drive which is a SMR 5900rpm drive. Rather interesting to see the large level of similarities between the products – including the price.
Testing was performed on my new workstation during the course of the storage commissioning tests. As the two units tested exhibited very similar performance, the results for just one unit will be presented.
Right off the bat, the drive posts an impressive outer-diameter sequential read rate of 227.4MB/s, averaging 185.2MB/s. The curve is rather smooth, suggesting some sophisticated recording scheme which keeps all tracks at the same areal density (rather than the stepped multi-zone recording of old, or the “wavy” adaptive density schemes). The drive does show three conspicuous dip zones, which suggests these areas may be reserved for slipping/reallocations or drive operating firmware.
Sequential write performance was much the same, with the same characteristic shape, although the access time was a bit scattered suggesting some firmware-optimization behaviour (i.e. reporting writes cleared before they actually happen). But that’s hardly anything new as most drives do this to some degree anyway. However, the proportion of higher-access times seems a little unusual.
This behaviour extended to the random access tests. While hard drives don’t break any IOPS records, it seems clear that the write IOPS are higher than read IOPS. No big surprise as many drives are optimized this way. The big surprise was random writes where the maximum access time spiked to 2.5 seconds – this doesn’t seem particularly befitting for a surveillance drive, but then again, perhaps surveillance recording is sequential by nature.
Extra tests were performed, although they’re not particularly relevant in general. No file benchmarks were done with HD Tune Pro, unfortunately, however we do have CDM and ATTO to go on.
CDM reports performance a little higher than expectations based on the HDTune Pro results. The results are particularly good for a hard drive, especially when compared to older drives. However, as usual, small block accesses are much better handled by solid state storage.
It was so unexpectedly good that I re-ran the test a few times including with no overlapping I/O and with a larger transfer length. The performance remained similar, with full rates at 32kB, so it was not likely to be a result of caching.
At least it wasn’t the very inconsistent results as with the SMR Archive drive. It seems this drive may well be PMR.
A full test was performed with H2testW and no data integrity errors were reported. Note that the error on the left is normal as H2testW calculates the remaining space at the beginning and this is not actually available due to filesystem growth during testing as files are being created.
As sourced, the drive had clean values and no unusual indications.
After extensive testing, the drive remained healthy, however the Hardware ECC Recovered attribute 0xC3 was an unusually low figure (discussed in the next section). The drive maintained a 44 degree temperature under load while being cooled in front of a fan, so it’s quite a warm drive with a claimed 9W average operating power and 8W idling average (which is higher than green drives of the past, or even the archive which has an average operating of 7.5W and idle average of 5W). That’s expected from the higher spindle speed. The time to ready is also around 30 seconds, and it’s noticeably slow to get up to speed when first powered up or awoken from sleep. For those who don’t sleep their drives, that’s no big deal.
The (Possible) Wrinkle in the Plan?
While the drive seems to be everything I had wished for, especially at the price point, all may not be quite perfect. Judging from the SMART data, the low value for Hardware ECC Recovered suggests that there may be a penalty from going to PMR at such high densities.
In the majority of the Seagate drives I’ve owned, the value for this figure sits at about 30 to 55 for most mainstream drives, and about 100 to 120 for my Archive drives. The only exception was a pair of Seagate ST2000DL003 which had values around 5 to 6, one of which failed within about three years, developing slow I/O and 3975 reallocated sectors.
I suspect the SMART value denotes the relative margin between the read-signal and the point of failure where error correction is unable to restore the recorded data. I suspect low values indicate that the recovered signal is closer to the threshold of unrecoverability, although this is merely speculation.
As the SMART data is not standardized between manufacturers and models, and manufacturers are not proactive about informing consumers about exactly what the metrics mean, this is just my suspicion. However, it does make sense considering that just two years ago, SMR was absolutely necessary to achieve such densities in non-helium drives while meeting consumer expectations and that since then, no major breakthroughs in PMR technology have eventuated.
As a conservative person when it comes to new storage technologies, I lived through the LMR to PMR transition period. When LMR was reaching its technological limits, the largest drives reached 500Gb using five platter arrangements. The majority of mainstream user drives had up to three platters, so we saw capacities around 320Gb at the end of LMR’s life. At the time, newer 500Gb, 600Gb and 750Gb drives with PMR were being released but as I was skeptical of the reliability of the drives, I deployed a number of 320Gb drives in a NAS instead.
This ironically proved to be a bad choice, as the 320Gb drives operated so close to their margins that weak sectors developed over two of the eight drives I had, resulting in sporadic data loss that (when rewritten) did not result in reallocations, causing further data loss later on. The new drives using PMR that my friends embraced, interestingly enough, had no major issues and neither did my older 200 and 250Gb drives based on a less-dense platter arrangement.
I guess the same scenario might be playing out again. As a skeptic of the long-term reliability of helium based drives (and recognizing that helium needs to be conserved), I have resisted the move to helium. Another reason is cost. But as we push PMR to its limits, Seagate seemed to make a wise decision to back-off on density and move to SMR in the Archive series. But it seems that the market has not particularly taken to SMR drives with the same enthusiasm (even if they are inadvertently given them in external drives). Might Seagate be abandoning SMR in favour of pushing PMR even further – perhaps too far? I guess only time will tell.
While the Seagate Skyhawk Surveillance drive shares similarities with the older Seagate Archive drive in terms of most of its parts, the performance of the two drives are quite different. The Skyhawk’s faster 7200rpm spindle speed provides much improved sequential performance, and its layout seems to resemble a traditional PMR drive without any of the SMR-induced throughput variations and slow-downs experienced in real life. The Skyhawk is also cheaper, making the Archive drive practically irrelevant for most consumers. This drive is able to offer such a capacity without resorting to helium-based technology, which keeps the price affordable.
Because of the similarities, the drive (likely) has the same 6-platter, 12-head arrangement which may prove to be a reliability issue. Because of the spinning mass, it also takes about 30 seconds to come up from a spindle-stopped condition. That being said, my four Archive drives have clocked 18,000 hours each with nil failures, however, your mileage may vary. It’s also fairly heavy and runs relatively warm (testing was conducted with a fan directly in front of the drive bay). It misses out the centre mounting hole on both sides which may impact compatibility with some cases.
For the most part, the drive seems to be a good choice in general – it’s large, it’s fast, it’s consistent, and it’s relatively inexpensive. However, whether it proves to be reliable in the long run can only be assessed through the passage of time.