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

Why cant they use each heads output in parallel instead of reading from only one at a time?

I believe precise alignment to all heads / platters at the same time is impossible considering the temperature will change during operation.
 
....and the only way it would work efficiently would be a harddrive incapable of ever needing a defrag. As soon as data is placed out-of-order, as most data is, data rate would go back to one head read speeds.

....or, is data purposely placed 180 degrees apart so data can be read consecutively by the two heads :/
 
i dont know why, but i had thought that the data was striped across all the heads for some time now.


There is something like that going on, but I am not entirely sure how.

As data densities have increased, there is some checksumming and data protection going on internally on hard drives these days. That;s why ytou can have raw read errors reported by SMART, yte still ahve read your data properly.

Not quite sure how they do it though. It's possible that the heads just move at the same time, but one of the platters is used for parity.

So, the change with this tech would be that you have some independent motion. Essentially, it would be like having two hard drives in the same enclosure, sharing a single motor.
 
I believe precise alignment to all heads / platters at the same time is impossible considering the temperature will change during operation.
The % change per head has a tolerance that determines the minimum size of each set of magnetic poles.
The tolerance wont change if a new read/write method is employed. They are already subject to those temp variations.
Each head already reads and writes at very specific locations, if any head has a problem the drive fails.
Using more than one head at a time doesnt change the tolerances or required tolerances.
 
This person explained it (the alignment problem with temperature changes) better than me:

https://forums.theregister.co.uk/forum/1/2017/12/19/seagate_disk_drive_multi_actuator/#c_3378410

Because a head is servo positioned on the track it is reading or writing and the alignment between head and track on other platters is no where near good enough. For example at current densities a 1 degree C temperature differential between two aluminium disks would produce a thermal expansion differential about 6 tracks wide at 1 inch from the spindle.

I would speculate micro (probably piezo) actuators on each arm could accommodate errors and allow one head on each arm to servo onto the 'same' track. Would need a lot more electronics to achieve.
 
Last edited:

...I still feel that it could be done better by writing the data in such a way that opposing heads are 'naturally' placed inline with the next stripe of data. Exactly like raid 0 except the data would be precisely placed by simply alternating data write's to alternate heads. If you write data alternatively to head1 then head2 then the data will be read at twice the speed.

The servo logic to precisely place the heads already exist. Defragging such a drive would be a monster.
 
This product is the recipient of my coveted "Nope" award.
 
I don't see these panning out all that well. The main advantage to them is having faster speeds in the same size package. But I don't know where the package space is that much of a concern that this would be the answer. The costs for this are likely greater than 2 drives of half the capacity striped in a traditional fashion, and if you're concerned about space solid state is going dominate that market. These would help random IOPS by basically cutting the number of platters in half per disk, but the rotational latency will be the same. If IOPS were a concern you're probably buffering those requests with SSDs. The only thing I can think of is if you had a need for a ton of storage capacity, but were trying to keep the costs down. It would be data that is used infrequently enough that you don't want to pay for capacity to buffer it (Like databases) but still needs to be on demand (Not backups that are never being used again)

The best scenario that these might be aimed at is datacenters with virtualization. You don't need 90% of the data for a VM buffered, but you also never know when that data is going to be accessed. The random I/O is going to be important to you as the amount of capacity per drive is well above your I/O abilities. I've definitely noticed where "cold data" can be slow to access from 7K disks, like when you want to apply patches to a server and you need to have it read parts of the VM that haven't been accessed in a while. With say 12TB drives you could hold 2x the amount of VMs, but in reality you'll still end up with 2x as many drives you needed for capacity just so you can cover the I/O needed to actually get at that data. The drives are cheap, but the SANs that hold them are not. This would probably end up being slightly cheaper than buying the drive shelf needed to hold the drives as I'm sure the cost per drive is going to be higher.
 
I agree but a compensation method on each head is a fairly straight forward addition.
A general compensation applied to the head actuator (as already performed) and a tiny offset for each head applied with a tiny amount of heat at a thermal junction in the metal holding each head.
Or the full compensation for each head could be applied at the junctions.

Or perhaps even simpler.
On each seek mechanism, test each heads thermal expansion and adjust so they all respond the same before fitting.
If there is a different coefficient of expansion for the different platter positions, incorporate that on the appropriate heads as well.
 
I think it would be an awesome design implemented on a Solid State drive. Would M.2 Raid 0 performance be faster than a single controller?
 
SSDs are very much RAID0 internally. This is why larger drives with more NAND usually perform better than small drives.
 
Solid scrape Connor Chinook. Yeahverily!
Next up? Massively parallel punched paper tape...
And just when player piano rolls were starting to look obsolete.
 
More magnets to salvage from a single broken drive! I replace broken Seagates all the time so it's good news for me :)
 
Solid scrape Connor Chinook. Yeahverily!
Next up? Massively parallel punched paper tape...
And just when player piano rolls were starting to look obsolete.

Also the return of 8 Track cartridges as "8x Redundant Storage"!!!
 
This coming from a hard drive manufacturer named Seagate which carries a novelty label known for high number of HDD failures?

Pass.
 
Aren't SSDs already internally RAID-0ed by design?
Anything in parallel is.
Dual channel or quad channel ram for example.
Its not best thought of as RAID hard drives because that uses different spindles which result in different times to first access. It has more in common with dual channel memory.
 
Last edited:
... and since it is Seagate, this will also cut reliability in half.

and thats bad when they are craptastic drives to begin with, seagate and wd green/blue drives... the bane of the IT world
 
... and since it is Seagate, this will also cut reliability in half.

Says the person who is clearly thinking only of the cheap consumer drives. Enterprise and Pro level of drives are just fine. Oh, and I have 2 x 1TB drives in my home computer giving me no issues as well as a 2TB drive in my other computer, still going strong, all Seagate. Yes, they have some drives that have issues but, not everything.
 
if one actuator fails they should be able to still use the other. just at old performance speeds.


1/2 of the drive would not be readable since you can't seek with a stuck actuator. I highly doubt there would be any way to lock the 2 halfs back together forming a normal single actuator system.
 
Last edited:
  • Like
Reactions: Nenu
like this
Aren't SSDs already internally RAID-0ed by design?

Not directly. The data is scattered all over the drive on purpose so as to wear down the NAND flash cells evenly. The controller is then able to access all those random locations at the same time. So you get speed gains from parallel access. Which is what RAID0 is doing, but the SSD wasn't done with the intention of RAID0, it just happens to be similar. This is why we don't "de-fragment" an SSD drive, it's fragmented on purpose, and offers no performance increase.
 
  • Like
Reactions: DPI
like this
I'm not gonna rain on their parade, I think it's cool to try new things even if it is a failure.

I wish they would bring back the SSD+HDD drives that had a real SSD and then an HDD mapped all into the same partition. It was like the first 128GB of the partition were on SSD and then the remaining TB or whatever were on HDD.

The best method would be to make the drive be able to talk to the OS and let it dynamically move files around on the drive so that most used things or important files are kept on the SSD section. Like Apple's Fusion setup, only with 1 single drive. The current SSHD hybrid drives (with a measly amount of flash) are not "OS aware", and just have to guess on what to move to the SSD cache.
 
I'm not gonna rain on their parade, I think it's cool to try new things even if it is a failure.

I wish they would bring back the SSD+HDD drives that had a real SSD and then an HDD mapped all into the same partition. It was like the first 128GB of the partition were on SSD and then the remaining TB or whatever were on HDD.

The best method would be to make the drive be able to talk to the OS and let it dynamically move files around on the drive so that most used things or important files are kept on the SSD section. Like Apple's Fusion setup, only with 1 single drive. The current SSHD hybrid drives (with a measly amount of flash) are not "OS aware", and just have to guess on what to move to the SSD cache.

Actually the best design would be to have all the memory built into the CPU and eliminate the need for External drives all together. Although I think were a long way from that.
 
1/2 of the drive would not be readable since you can't seek with a stuck actuator. I highly doubt there would be any way to lock the 2 halfs back together forming a normal single actuator system.

Yeah, I think the only way to do this "properly" and by that I also mean inefficiently and not worth the effort on spinning media, would be to have two FULL sets. That would basically give you RAID1+0ish capability, though if anything happens to the platters themselves you're still screwed. The cost would also not be worth it. There would also need to be a fairly involved controller to handle the multiple write timing within the same platter. Yuck...
 
Last edited by a moderator:
1/2 of the drive would not be readable since you can't seek with a stuck actuator. I highly doubt there would be any way to lock the 2 halfs back together forming a normal single actuator system.

I could be misunderstanding how it works then. I thought there'd be two identical actuators that both read the same accessible data; one on the right, one on the left, outside to inside.
So as the platters spin you can read as it passes the first actuator and again as it passes the 2nd.

Is this not the case?
 
Last edited:
No this version (of the several decades old idea) has 2 actuators on a single pivot point. The top actuator access the top 1/2 of platters.
 
Last edited:
Actually the best design would be to have all the memory built into the CPU and eliminate the need for External drives all together. Although I think were a long way from that.

We are getting there with nVME and eMMC memory. eMMC is built right in to the motherboard.
 
Back
Top