Future Seagate HDDs will begin including two separate sets of actuator arms that can operate independently in order to double read/write speeds:
Seagate's multi-actuator technology is a simple concept, and the idea certainly isn't new. In fact, the company has already developed drives with multiple actuators in the past, but they weren't economically viable due to higher component costs.
Most HDDs read and write data to and from multiple platters. For instance, Seagate's largest drives wield up to 8 platters and 16 heads. The heads, which are connected to the end of an actuator arm assembly, read and write data from both sides of each platter.
Unfortunately, those 16 heads are all aligned on the same arm, which means they all move in unison. Simultaneously aligning all the heads on all the platters isn't possible because of the incredibly thin data tracks on the platters, so only one of the heads is actively reading or writing data at any given time. That limits read/write throughput and performance with randomly accessed data.
Seagate's new design uses two sets of actuator arms that operate independently. Each carries eight heads. That means the drive can read or write from two heads at once, provided they are attached to different actuator arms. The drive can respond to two commands in parallel and one head can read while another writes, or both can read or write simultaneously.
Related Stories
The Register asked Seagate's Director of Technology Strategy and Product Planning Jason Feist about the company's plans to use multi-actuator technology in upcoming hard disk drives. Seagate insists that the technology can double input/output operations per second (rather than increasing it by, for example, 1.8x), and says that customers have validated the concept:
Howard Marks, founder and chief scientist at Deep Storage Net said: "We've had drives with 2 positioners before (IBM 3380 - one set of heads were dedicated to inner tracks, the other to outer tracks). That was back in the day of linear voice coils so they came from opposite sides of the 14-inch platters."
He identifies a software issue with Seagate's multi-actuator single pivot design: "Most storage software including logical volume managers and file systems, are built with the knowledge that a disk drive can only have it's heads in one place at a time and their queuing logic may mismatch with the multipositioner logic." This means: "It may not double throughput for large I/Os."
It could get close though, as "I understand that Seagate is going to make these look like 2 logical drives via a driver. That should solve #1 above and let systems get 1.8-1.9X IOPS."
(Score: 1, Interesting) by Anonymous Coward on Wednesday December 20 2017, @10:22AM (10 children)
Let us double the number of parts to fail!
Honestly, double actuators for the same track would make more sense. That way if you started having a head failure with one, you have a backup to offload your data before the drive fails completely.
This way, you just have more of a chance of losing half the drive, and most likely interleaved cylinders/sectors/tracks if it provides a throughput boost.
New hardware designs are definitely not always better than old ones.
(Score: 0) by Anonymous Coward on Wednesday December 20 2017, @10:46AM (1 child)
You make a good point that it /could/ lead to longer living drives on top of the better througput. Unfortunately I can't help but see this as a late in the game attempt by Seagate to remain relevant. If you care about perfoance in a drive you're likely picking a Samsung or Intel SSD over the lackluster Seagate hybrids.
This tech is seriously old. It was once used with 2 separate interfaces so you could deliver content to webservers through read-only arms. Pretty sure that died a quick death.
Conspiracy theory - "BONUS" they could set the SMART data however they like so you are only warned about a bad read/write element outside of warranty, as the failure can remain "hidden" for longer. **yay Seagate**
(Score: 3, Informative) by takyon on Wednesday December 20 2017, @02:46PM
Not sure I would tie this so closely to Seagate. Once Seagate does something, WD is sure to follow, and vice versa. Even Toshiba still has skin in the game, miraculously.
WD/HGST moved to helium-filled drives, Seagate followed, and later Toshiba [anandtech.com]. Both companies had been working on it for years.
All three companies have moved past 5 platters and back into the territory of 7-9 platters. We could see 12 platters on a glass substrate [soylentnews.org] in 3.5".
Seagate is moving to HAMR, WD is moving to MAMR, which is similar.
Seagate is using shingles (although not in all drives), even with its disadvantages, and so is WD [storagereview.com].
Both companies are working on two-dimensional magnetic recording (TDMR), which should boost capacity by a few percent.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 0) by Anonymous Coward on Wednesday December 20 2017, @10:48AM (1 child)
If I understand the concept correctly, each individually operated actuator + set of heads can cover the entire drive surface anyway, which covers your scenario of a single head failure. If anything, if designed correctly, I see doubling of parts as creating redundancy inside the drive, not adding more single points of failure. The additional read/write throughput is a bonus. If you lose a head on today's drives, you lose access to the data on what, once side of an entire platter? This design fixes that issue. Not that I'm sure if it's even a common failure case.
(Score: 5, Informative) by WizardFusion on Wednesday December 20 2017, @10:59AM
Not according to the animation.
One actuator arm covers the top half of the platters, and the other arm covers the bottom half. If you lose a head, you still lose access to the platter.
(Score: 3, Interesting) by shrewdsheep on Wednesday December 20 2017, @02:35PM
OTOH the force needed to move each arm is halving (or so). So maybe that is a good thing.
(Score: 2) by frojack on Wednesday December 20 2017, @07:07PM (2 children)
The source of the underlying problem was mentioned:
It almost leads one to believe the tracks are there on the drive before the build it.
I thought those things were laid down as needed, and the system remembered just how far to move the arm to find it again.
It would seem that you could always seek to a given physical track and create other tracks would align under the other heads.
These x tracks could be read and written in parallel.
The system would then only need to remember that when you wanted track 342, you need to seek head one to 338, and then read with head 5 (as an example), but also with heads 1-n would also be aligned on tracks that were logically (but not necessarily physically) consecutive tracks, and could be accessed in parallel.
No, you are mistaken. I've always had this sig.
(Score: 2) by bob_super on Wednesday December 20 2017, @07:23PM
> when you wanted track 342, you need to seek head one to 338, and then read with head 5
When you need to read from track 342, go to track 342 and read/write using all 8 to 16 heads, with the data split between platters, multiplying throughput. Bit interleave gives the fastest latency to first word, but byte or word interleaves would still provide the benefit.
(Score: 0) by Anonymous Coward on Thursday December 21 2017, @02:44PM
I always assumed the arms/heads laid down the tracks and that's why you can't have different heads on different arms serving the same track. Their opinions on where track #12 is would be different enough to not work well...
(Score: 2) by Immerman on Wednesday January 10 2018, @03:32PM (1 child)
How common is head failure as a cause of drive failure though? I tend to use drives to destruction, and almost all of them have failed in the "rapidly expanding regions of bad sectors" typical of surface failure. The remainder simply became completely non-responsive, disappearing even from the BIOS, which I assume indicates control board failure.
Not once have I seen the "regular stripes of unreadable data" that I would assume would be characteristic of a failed head.
That said, I would love to see multiple arms per platter for performance reasons, seems like you should be able to fit at least a few around the platters without even redesigning them.
The performance benefits of this seem... dubious, despite their in-house benchmark claims. It is after all effectively two mostly independent half-size drives sharing the same spindle and control board - and by it's nature each arm will have only half the throughput of a single integrated arm. So linear I/Os will have, at best, the same speed as they would with a single arm, while random I/O will only benefit if successive accesses fall on opposite semi-drives. A deep enough command queue should make that feasible to arrange, but I'm not sure how typical that demand would be.
Really, I can't see any way they'd see anything close to 2x the performance in real-world benchmarks unless they were comparing it to a drive with only half the number of platters.
(Score: 2) by Immerman on Wednesday January 10 2018, @03:41PM
Wait. Never mind, I missed:
> Simultaneously aligning all the heads on all the platters isn't possible because of the incredibly thin data tracks on the platters
I always assumed all the heads operated simultaneously so that you could read/write an entire cylinder at once. If you're only using one head at a time, then yeah, being able to use two heads at a time would double your throughput even for linear I/Os.
(Score: 1, Interesting) by Anonymous Coward on Wednesday December 20 2017, @05:00PM
The added IOPS (IO per second) is nice, but the driver is probably more to do with reducing device test time.
Formatting can be done in half the time if you are doing it twice as fast. When drives are getting into multiple TB range, this is a surprisingly big savings.
(Score: 4, Interesting) by Grishnakh on Wednesday December 20 2017, @06:10PM (1 child)
First, I didn't read TFA, so I don't know if this drive is designed so that the heads act in parallel on the same data, or different data. If it's different data (meaning, the data is not interleaved between the upper platters with their R/W heads and the lower platters with their separate set of R/W heads), then all they've really done here is stick two HDs together into a single package, and then hide this fact with the HD firmware. If it's the same data (it's interleaved for better performance on a single R/W request), then what they've effectively done is create a RAID0 system in a single HD package. In the first case, you'll get better performance for applications where different processes are reading or writing to different areas of the disk, but without direct control, you may or may not realize the benefit, depending on whether the HD firmware happens to put the data for two active processes on the different sides of the drive. In the second case, you just had a RAID0 system, with all the disadvantages that come with RAID0, though with them sharing the same spindle, theoretically the reliability should be a bit higher than with separate spindles. But a problem on either side of the drive, with either set of heads, means all your data is gone, and the probability for this is higher since there's two independent head assemblies.
This would have made more sense I think if the two head assemblies were duplicates of each other, offering a level of redundancy (as long as a head failure doesn't wreck the platters), in addition to being able to speed up access to data no matter where it is on the platters. The way it is here, it's highly dependent on the firmware being able to intelligently place the data on the platters, or on the OS communicating to the HD where it should (which isn't going to happen).
(Score: 2) by frojack on Wednesday December 20 2017, @07:16PM
I used to know how stuff was laid out between platters, but that was decades ago. I have no clue any more.
I think you and I are on the same track (bad pun) on this. See my post [soylentnews.org] above.
Build the tracks on the stacked platters on the fly, and then just due seek to track A on platter one, and track A2, A3, An would be aligned on platters 2,3,n and could be written/read in parallel with one arm.
No, you are mistaken. I've always had this sig.
(Score: 3, Interesting) by bob_super on Wednesday December 20 2017, @07:17PM (1 child)
> Simultaneously aligning all the heads on all the platters isn't possible because of the incredibly thin data tracks on the platters
I thought all those heads could read-write at the same time. I'm genuinely surprised they don't
Finding a way to force that platter-to-platter alignment, and allow all heads to run at the same time, would multiply throughput by the number of heads with no additional BOM cost and no impact on reliability (if feasible at all). Not sure what the constraints are, since alignment can be found on each platter.
If you're gonna put more actuators, how about (as already suggested) putting twice or three times as many heads with up to three sets of arms at 120 degrees around the disk? Sure it'd be expensive, but provide much better random access times and still fail pretty gracefully (which the SSDs don't).
(Score: 0) by Anonymous Coward on Thursday December 21 2017, @08:41AM
>fail pretty gracefully
unless there's a head crash
(Score: 0) by Anonymous Coward on Wednesday December 20 2017, @10:10PM
http://www.instructables.com/id/Hard-Drive-Speaker/ [instructables.com]