Tuesday, December 19th 2017

Seagate Bringing RAID 0 Performance to Single HDD via Multi-Actuator Tech

Seagate may currently be one of the first tech companies in storage conversations due to their upcoming HAMR (Heat-Assisted Magnetic Recording) technology, which should enable 20 TB HDDs by 2020. And even though HDDs are better known for their high areal density and price/storage ratio, which is only bound to increase through the usage of HAMR, Seagate knows that areal density is hardly one of the principal bottlenecks in HDD technology. The bottleneck, as it usually is, is speed.

HDDs have an old design philosophy by now, where an actuator arm moves read/writing heads in parallel across the surface of the disk - nowadays, there are usually two heads per platter (one on the upper side, and one on the underside of it). As you might image, a given head can either read or write at one point in time - and all heads move in tandem, with different heads reading or writing across the multiple spinning discs that constitute the hard drive. This, however, means that HDD reading efficiency is lost - due to how small the 1s and 0s are on HDDs, only one head can be moved to an exact data path, with all the others moving with it, spending power and increasing the load on the actuator for nary a speed gain or minimal workload.
Seagate's solution is ingeniously simple (for us laymen looking from the outside in, naturally). Seagate is planning to install not one, but two actuator arms in HDDs, which would work entirely separately. Naturally, still only one head in each actuator would be able to read or write at any point in time, but nevertheless, this implementation would mean that there can be two HDD sections being operated on at the same time, effectively allowing for double the read and write speeds. This isn't much too different from RAID 0 configurations, really - where data from a single HDD is mirrored across two HDDs, which guarantees that the heads of each HDD can both be reading/writing the same data, thus doubling read and write speeds. This implementation by Seagate could either allow for an intra-disk RAID 0 array - or simply for a bolster to an HDD's ability to simultaneously read and write, without any performance loss - if the data you're reading and writing are on the platters supported by different actuators, naturally.
"Seagate Hard Drives are about to embark on a new exponential growth in capacity with the introduction of EAMR technologies like HAMR [Heat-Assisted Magnetic Recording]," said Aaron Ogus, Microsoft Azure Storage Architect. "In most datacenter applications the additional capacity gains cannot be effectively utilized without improvements in device IO capacity. The dual actuator technology helps unlock additional IOPS [input/output operations per second] and allows cloud providers to make effective use of the new capacity gains."
Seagate says the drives could use SAS, SATA, or NVMe interfaces, which is a boon for all kind of customers, be them enterprise or consumers. Closing this article up, this editor would like to say that these are probably the most interesting times in HDD's life in decades now. The medium really hasn't evolved all that much since its inception. However, we are now on the verge of seeing the two greatest HDD manufacturers, WD and Seagate, coming out of the gates with different technologies for the medium. HAMR and MAMR (Microwave-Assisted Magnetic Recording) can either be great for customers, with increased product differentiation according to a users' needs, or might result in one company struggling much more than the other, considering implementation difficulties of each technology. Now, with the double-actuator design that Seagate is presenting, it seems we'll have another point of differentiation sometime in the future - though this seems more like a safe bet for all manufacturers to adopt, should it prove worthy of the investment. Fun times. Source: Seagate Blog
Add your own comment

27 Comments on Seagate Bringing RAID 0 Performance to Single HDD via Multi-Actuator Tech

#1
TheLostSwede
This would work well for sequential data once again, but not for random data. The two actuator arms would also have to be synced for some types of data, or this could in fact end up being slower than a regular hard drive. I guess it all comes down to how good they are at designing a firmware that can take advantage of this, but it seems like it's going to have an equal amount of downsides as it has upsides.
Posted on Reply
#2
R-T-B
TheLostSwede said:
This would work well for sequential data once again, but not for random data. The two actuator arms would also have to be synced for some types of data, or this could in fact end up being slower than a regular hard drive. I guess it all comes down to how good they are at designing a firmware that can take advantage of this, but it seems like it's going to have an equal amount of downsides as it has upsides.
What if the arms were simply positioned at opposing ends of the disk platter?
Posted on Reply
#3
HopelesslyFaithful
TheLostSwede said:
This would work well for sequential data once again, but not for random data. The two actuator arms would also have to be synced for some types of data, or this could in fact end up being slower than a regular hard drive. I guess it all comes down to how good they are at designing a firmware that can take advantage of this, but it seems like it's going to have an equal amount of downsides as it has upsides.
I dont get what your saying.

If you have 2 independent heads for reading that means these would be better for random reads assuming the random reads were on top and bottom platters. This goes the same for sequential. It is only faster if data is on the top and bottom. If all the data is on the top than it isn't any faster.

Basically this doesn't function like a RAID0 (unless they design it that way). It would function like a JBOD. You could design it as either one but i was reading it as a JBOD not as a RAID0....its smarter and safer to have it run as a JBOD because if the bottom actuator died....the top data is still readable. If it is actually striping data that means if 1 actuator dies everything is lost vs if it ran as a JBOD.

I would rather see it as a JBOD personally.

Also i have been wondering how long it would take them to do this....I was asking why dont be have independent arms for each platter back in the early 2000s
Posted on Reply
#4
TheLostSwede
HopelesslyFaithful said:
I dont get what your saying.

If you have 2 independent heads for reading that means these would be better for random reads assuming the random reads were on top and bottom platters. This goes the same for sequential. It is only faster if data is on the top and bottom. If all the data is on the top than it isn't any faster.

Basically this doesn't function like a RAID0 (unless they design it that way). It would function like a JBOD. You could design it as either one but i was reading it as a JBOD not as a RAID0....its smarter and safer to have it run as a JBOD because if the bottom actuator died....the top data is still readable. If it is actually striping data that means if 1 actuator dies everything is lost vs if it ran as a JBOD.

I would rather see it as a JBOD personally.

Also i have been wondering how long it would take them to do this....I was asking why dont be have independent arms for each platter back in the early 2000s
Since when have hard drives been good for random data? Even this isn't going to help random access scenarios, but it would potentially help speed up sequential data in the same way RAID 0 would. Sure, if the random data is written to both parts of the drive, we might go from less than 1MB/s to 1.5MB/s, but that's hardly going to revolutionise hard drive storage.

I never said it would be like RAID 0, that was the author of the article.

The problem with individual arms is fitting all the actuators, as they take up a fair amount of space and they also add a fair amount of cost and complexity, which is most likely why no-one's done it.

R-T-B said:
What if the arms were simply positioned at opposing ends of the disk platter?
You want a return to 5.25" drives? That would most likely not fit in current form factors.
Posted on Reply
#5
Ebo
For server parks I can see the benefit, but for consumers in general( like you and me) absolutely nothing, just more moving parts which can fail due to density between plates.
Posted on Reply
#6
R-T-B
TheLostSwede said:
You want a return to 5.25" drives? That would most likely not fit in current form factors.
If it improves HDD performance?

Yes.

That aside, I would assume we are capable of making things a lot smaller than what we were able to do when we made 5.25" drives.

Of course eventually pesky physics get in the way...
Posted on Reply
#7
Nehemoth
TheLostSwede said:
Since when have hard drives been good for random data? Even this isn't going to help random access scenarios, but it would potentially help speed up sequential data in the same way RAID 0 would. Sure, if the random data is written to both parts of the drive, we might go from less than 1MB/s to 1.5MB/s, but that's hardly going to revolutionise hard drive storage.

I never said it would be like RAID 0, that was the author of the article.

The problem with individual arms is fitting all the actuators, as they take up a fair amount of space and they also add a fair amount of cost and complexity, which is most likely why no-one's done it.



You want a return to 5.25" drives? That would most likely not fit in current form factors.
I do undertand what @TheLostSwede is saying. Even if the actuators are as oppose end it would need a firmware that allow to divide the data in this new way, and for once I wonder if this wouldn't cause fragmentation.
Posted on Reply
#8
plåtburken
I am curious to know how hot the whole drive would be during full load.
Posted on Reply
#9
HopelesslyFaithful
TheLostSwede said:
Since when have hard drives been good for random data? Even this isn't going to help random access scenarios, but it would potentially help speed up sequential data in the same way RAID 0 would. Sure, if the random data is written to both parts of the drive, we might go from less than 1MB/s to 1.5MB/s, but that's hardly going to revolutionise hard drive storage.

I never said it would be like RAID 0, that was the author of the article.

The problem with individual arms is fitting all the actuators, as they take up a fair amount of space and they also add a fair amount of cost and complexity, which is most likely why no-one's done it.



You want a return to 5.25" drives? That would most likely not fit in current form factors.
You seem to be very special. You stated RAID0 does not affect random and that it most likely hurts it when it comes to HDDs. That is patently false but nice try back tracking and goal post shifting. RAID 0 should result in 1-2x random just like sequential reads/writes should be 1-2x depending on how the data is laid out for HDDs.

R-T-B said:
If it improves HDD performance?

Yes.

That aside, I would assume we are capable of making things a lot smaller than what we were able to do when we made 5.25" drives.

Of course eventually pesky physics get in the way...
100% agree. For servers I don't think size perse is a huge issue when it comes to large arrays and performance.

plåtburken said:
I am curious to know how hot the whole drive would be during full load.
I dont think it would be much worse. I maybe wrong but i would assuming the spinning results in most heat and not the actuator.
Posted on Reply
#10
TheLostSwede
HopelesslyFaithful said:
You seem to be very special. You stated RAID0 does not affect random and that it most likely hurts it when it comes to HDDs. That is patently false but nice try back tracking and goal post shifting. RAID 0 should result in 1-2x random just like sequential reads/writes should be 1-2x depending on how the data is laid out for HDDs.
I'm sorry, what? Nowhere in my first post does it say RAID, I think you need to learn to read or put on your glasses. The only one talking about this working like RAID-0 is the news poster, not me. Get your facts right before accusing people of things.

R-T-B said:
What if the arms were simply positioned at opposing ends of the disk platter?
Looks like good old Conner did it way back in the day, but it seems like it was never available to the general public, as I can't remember this ever having been marketed or sold.
https://en.wikipedia.org/wiki/Conner_Peripherals#Performance_issues_and_the_"Chinook"_dual-actuator_drive
Posted on Reply
#11
thesmokingman
TheLostSwede said:
This would work well for sequential data once again, but not for random data. The two actuator arms would also have to be synced for some types of data, or this could in fact end up being slower than a regular hard drive. I guess it all comes down to how good they are at designing a firmware that can take advantage of this, but it seems like it's going to have an equal amount of downsides as it has upsides.
Because you know more about storage than Seagate?
Posted on Reply
#12
trparky
thesmokingman said:
Because you know more about storage than Seagate?
Considering Seagate's reliability is shit, it appears even Seagate doesn't know data.

Seagate: cause your data? f**k it.

Friends don't let friends buy Seagate.
Posted on Reply
#13
HopelesslyFaithful
TheLostSwede said:
I'm sorry, what? Nowhere in my first post does it say RAID, I think you need to learn to read or put on your glasses. The only one talking about this working like RAID-0 is the news poster, not me. Get your facts right before accusing people of things.



Looks like good old Conner did it way back in the day, but it seems like it was never available to the general public, as I can't remember this ever having been marketed or sold.
https://en.wikipedia.org/wiki/Conner_Peripherals#Performance_issues_and_the_"Chinook"_dual-actuator_drive

TheLostSwede said:
This would work well for sequential data once again, but not for random data. The two actuator arms would also have to be synced for some types of data, or this could in fact end up being slower than a regular hard drive. I guess it all comes down to how good they are at designing a firmware that can take advantage of this, but it seems like it's going to have an equal amount of downsides as it has upsides.
either way what you said was total BS. So keep trying to save face buddy :roll:
TheLostSwede said:
Since when have hard drives been good for random data? Even this isn't going to help random access scenarios, but it would potentially help speed up sequential data in the same way RAID 0 would. Sure, if the random data is written to both parts of the drive, we might go from less than 1MB/s to 1.5MB/s, but that's hardly going to revolutionise hard drive storage.
still goal post shifting to save face...and making things up as you go. :laugh:

If my server wasn't busy ATM I could run a simple test and show you are making things up but i have to many things running ATM and not enough open slots to play around...all my bays are full on my Norco 4224 TT...well minus one...because its dead :shadedshu:
Posted on Reply
#14
bonehead123
Well, I have 2x Seagate SSD's that have been running strong for over 2 years now, so obviously their no-moving-parts division is a bit more talented that their spinner department is, hahahaha :D
Posted on Reply
#15
HopelesslyFaithful
bonehead123 said:
Well, I have 2x Seagate SSD's that have been running strong for over 2 years now, so obviously their no-moving-parts division is a bit more talented that their spinner department is, hahahaha :D
the last update from back blaze IIRC said that seagate drives are now in the "normal" range of reliability. They were real trash during the floods where they sold sub par units while the demand was super high and supply was low.
Posted on Reply
#16
Kursah
Only going to say this once, play nice or earn reply bans and infraction points.

Thanks! :toast:
Posted on Reply
#17
lexluthermiester
TheLostSwede said:
I'm sorry, what? Nowhere in my first post does it say RAID, I think you need to learn to read or put on your glasses. The only one talking about this working like RAID-0 is the news poster, not me. Get your facts right before accusing people of things.



Looks like good old Conner did it way back in the day, but it seems like it was never available to the general public, as I can't remember this ever having been marketed or sold.
https://en.wikipedia.org/wiki/Conner_Peripherals#Performance_issues_and_the_"Chinook"_dual-actuator_drive

Yes, this. After being bought out, there were prototypes for a quad actuator design. There is enough room for them on platter. It's a shame no one has done this yet in the enterprise and consumer space.
Posted on Reply
#18
FordGT90Concept
"I go fast!1!11!1!"
About time. When I realized they weren't RAID0ing the platters, I was very disappointed. This though, this looks complicated.
Posted on Reply
#19
lexluthermiester
FordGT90Concept said:
This though, this looks complicated.
Given the advances made in the 20ish years since Conner tried this concept out(good grief, has it really been that long?), it's likely not as complicated as you might think. Seems a very feasible way to speed up current HDD's without driving up the cost too much.
Posted on Reply
#20
_JP_
GIMME QUAD-ACTUATOR RAID60 64GB/512MB 15Krpm SAS SSHD NOW!! :laugh:
Posted on Reply
#21
Mirkoskji
Can someone explain why, when an hdd has two platters, can't they be arranged in a raid 0 configuration? I mean it seems pretty easy, reading heads are already synced. A dual platter hdd could have double the sequential speed of a single platter. Thanks
Posted on Reply
#22
Sasqui
HopelesslyFaithful said:
I would rather see it as a JBOD personally.
Well, you can either get speed via Raid 0, or more disk space via JBOD, but not both. In either case, both have the downside of being twice as likely to fail, bringing ALL of the data on the disks down with it.

I think it's a stupid idea. The only plus side is not having two physical hard drives.
Posted on Reply
#23
lexluthermiester
_JP_ said:
GIMME QUAD-ACTUATOR RAID60 64GB/512MB 15Krpm SAS SSHD NOW!! :laugh:
Drop the SAS part and I'm in!
Mirkoskji said:
Can someone explain why, when an hdd has two platters, can't they be arranged in a raid 0 configuration? I mean it seems pretty easy, reading heads are already synced. A dual platter hdd could have double the sequential speed of a single platter. Thanks
You would not be incorrect.
Posted on Reply
#24
natr0n
More parts = more failures.
Posted on Reply
#25
lexluthermiester
natr0n said:
More parts = more failures.
Unless reduntancy measures are properly built in. For example, if one set of heads fail, the controller can lock the failed head armature and send an alert of the failure. The remaining heads can then be used to transfer the data before removal. If you think about it, that is a greatly useful capability.
Posted on Reply
Add your own comment