Do the actuators read the same platters? If not then you will only really get a performance increase in Random Reading and Writing, and not Sequential.
Why would it? Allocating the data to ideal/aligned locations across all platters can be done just as well with a single actuator, after all. I guess this would make that job easier as you'd be less reliant on being able to find empty sectors in a "stack" across platters, but at least for reads, I don't understand how this would increase max theoretical speeds whatsoever. Might have a minor impact in real life due to twice the heads for finding fragmented data, but that's about it unless there's something we're missing.
Not if it stripes the data RAID-0 style, putting every other block on the other platters (or stripes).
You could even have internal RAID-1, although that wouldn't carry much water considering you have to replace the entire drive to fix one actuator - plus other failures could knock both out.
I've got a 2TB drive that has read 170MBps sequential. The random sucks. Once you remove random from the equation, either with a program like Primocache or similar they're quite livable.
An old SSD with 170MBps read/write speeds and 80MBps random feels incredible compared to mechanical drives.
Seagate has no issues bringing forward a drive as they purchased Conner in the 1990's as they got most of the patents around this idea. Western Digital had to wait for these patents to expire before moving forward (they have been expired for several years now). It would be odd to see Western Digital release a dual actuator drive before Seagate given history.
As for power consumption, the saving between two drives likely stems from there only being one motor spinning the platters and common PCB. I suspect that from a performance/watt perspective, these drives will be major winners.
We've had electric cars in the early 1900's. But they weren't very feasible then. Due to several factors, they are now. Same with this idea. And probably 20 years from now when we get technology into everybodys hands that is in some labs right now but just not feasible for large scale deployment. That is how technology has always been. This isn't some big revelation.
All the innovation on this tech probably comes from the HGST side that they bought sometime earlier this decade. HGST was first to develop Helium hdds for the data center and enterprise market. WDC's strength had been the mobile and desktop markets.
Since not every failure point is double like it is in RAID 0, no. But since there are more failure points than in standard drives, it is likely that the failure rate is higher (unless they do more binning and tuning). Eventually, it will probably be pretty much identical, unless you purchase quantities in the thousands or hundreds of thousands.
failure, i.e. head crash, would likely be the same, since there are the same number. the only active part that is doubled, may be, is the voice coil(s). what proportion of HDD failures are the VC? or are there more extra active parts?
To an extent, the performance of dual head (yup, just like my VCR) drives can be simulated by building a two drive RAID 0 array with an admission that there are obviously going to be some differences. I'm not saying this isn't worth doing, but I do worry that there isn't a lot of performance gain left on the table for mechanical drives even with a second actuator when compared to solid state solutions. I guess there's a cost per capacity advantage, but it looks like there is a certain amount of erosion of that advantage as SSD prices have fallen quite a bit and mechanical drive research looks like its lagging a bit.
Considering you can get a 2TB 2.5" SATAIII SSD drive for under $250 lately and it's both just as fast or faster AND only due to come down in price, the argument makes sense for personal computing.
But there are two places these drives can still win. Price per terabyte, especially in combination with terabytes per cubic inch. If a 12TB model sells for $1200, its single 3.5" bay is less than the three 2.5" bays needed for $2100 worth of 4TB SATAIII SSDs. That makes these still worthwhile in datacenter uses.
Regarding their first TCO benefit, it seems like shared components such as the case, ceiling and motor base assembly can't contribute much at all to the total cost, otherwise an 8TB wouldn't cost $230 while the 4TB costs $120.
Doing the math on that tells me that 4TB worth of platters costs $110 and the case, ceiling and motorbase assembly only costs about $10.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
23 Comments
Back to Article
olafgarten - Friday, March 8, 2019 - link
Do the actuators read the same platters? If not then you will only really get a performance increase in Random Reading and Writing, and not Sequential.crimsonson - Friday, March 8, 2019 - link
Hmmm. Maybe. If both actuators are used during the same write, sequential read should also be affected. Similar to RAID 1?Valantar - Friday, March 8, 2019 - link
Why would it? Allocating the data to ideal/aligned locations across all platters can be done just as well with a single actuator, after all. I guess this would make that job easier as you'd be less reliant on being able to find empty sectors in a "stack" across platters, but at least for reads, I don't understand how this would increase max theoretical speeds whatsoever. Might have a minor impact in real life due to twice the heads for finding fragmented data, but that's about it unless there's something we're missing.GreenReaper - Friday, March 8, 2019 - link
Not if it stripes the data RAID-0 style, putting every other block on the other platters (or stripes).You could even have internal RAID-1, although that wouldn't carry much water considering you have to replace the entire drive to fix one actuator - plus other failures could knock both out.
Xajel - Friday, March 8, 2019 - link
Nope, one actuator serve half of the platters while the other serves the other half.Stormprobe - Friday, March 8, 2019 - link
I assume this feature will be SAS specific since SATA is half-duplex.AWDSYCO - Friday, March 8, 2019 - link
SAS only0ldman79 - Saturday, March 9, 2019 - link
A little smarts on the controller itself and you just split the file over multiple platters.SSD splits the data over multiple chips, otherwise we'd be limited to the read rate of a single chip, which generally isn't that impressive.
0ldman79 - Saturday, March 9, 2019 - link
Also, sequential isn't that terrible.I've got a 2TB drive that has read 170MBps sequential. The random sucks. Once you remove random from the equation, either with a program like Primocache or similar they're quite livable.
An old SSD with 170MBps read/write speeds and 80MBps random feels incredible compared to mechanical drives.
boozed - Saturday, March 9, 2019 - link
Isn't that where you want the performance improvements with spinning rust?Gasaraki88 - Monday, March 11, 2019 - link
This will increase sequential performance also because this is essentially RAID 0 in one drive.Kevin G - Friday, March 8, 2019 - link
What is old is new again. Conner had this technology a long time ago:http://mydatarecoverylab.com/history-repeating-mul...
Seagate has no issues bringing forward a drive as they purchased Conner in the 1990's as they got most of the patents around this idea. Western Digital had to wait for these patents to expire before moving forward (they have been expired for several years now). It would be odd to see Western Digital release a dual actuator drive before Seagate given history.
As for power consumption, the saving between two drives likely stems from there only being one motor spinning the platters and common PCB. I suspect that from a performance/watt perspective, these drives will be major winners.
Death666Angel - Friday, March 8, 2019 - link
We've had electric cars in the early 1900's. But they weren't very feasible then. Due to several factors, they are now. Same with this idea. And probably 20 years from now when we get technology into everybodys hands that is in some labs right now but just not feasible for large scale deployment. That is how technology has always been. This isn't some big revelation.vishnumrao - Friday, March 8, 2019 - link
All the innovation on this tech probably comes from the HGST side that they bought sometime earlier this decade. HGST was first to develop Helium hdds for the data center and enterprise market. WDC's strength had been the mobile and desktop markets.29a - Friday, March 8, 2019 - link
Will these drives have a higher failure rate similar to RAID 0?Death666Angel - Friday, March 8, 2019 - link
Since not every failure point is double like it is in RAID 0, no. But since there are more failure points than in standard drives, it is likely that the failure rate is higher (unless they do more binning and tuning). Eventually, it will probably be pretty much identical, unless you purchase quantities in the thousands or hundreds of thousands.FunBunny2 - Friday, March 8, 2019 - link
failure, i.e. head crash, would likely be the same, since there are the same number. the only active part that is doubled, may be, is the voice coil(s). what proportion of HDD failures are the VC? or are there more extra active parts?PeachNCream - Friday, March 8, 2019 - link
To an extent, the performance of dual head (yup, just like my VCR) drives can be simulated by building a two drive RAID 0 array with an admission that there are obviously going to be some differences. I'm not saying this isn't worth doing, but I do worry that there isn't a lot of performance gain left on the table for mechanical drives even with a second actuator when compared to solid state solutions. I guess there's a cost per capacity advantage, but it looks like there is a certain amount of erosion of that advantage as SSD prices have fallen quite a bit and mechanical drive research looks like its lagging a bit.GregBulmash - Friday, March 8, 2019 - link
Considering you can get a 2TB 2.5" SATAIII SSD drive for under $250 lately and it's both just as fast or faster AND only due to come down in price, the argument makes sense for personal computing.But there are two places these drives can still win. Price per terabyte, especially in combination with terabytes per cubic inch. If a 12TB model sells for $1200, its single 3.5" bay is less than the three 2.5" bays needed for $2100 worth of 4TB SATAIII SSDs. That makes these still worthwhile in datacenter uses.
ksec - Friday, March 8, 2019 - link
That has a potential 1GB/s in Raid 0. We seriously need 5Gbps or even 10Gbps Ethernet networking equipment price to drop.pixelstuff - Friday, March 8, 2019 - link
Regarding their first TCO benefit, it seems like shared components such as the case, ceiling and motor base assembly can't contribute much at all to the total cost, otherwise an 8TB wouldn't cost $230 while the 4TB costs $120.Doing the math on that tells me that 4TB worth of platters costs $110 and the case, ceiling and motorbase assembly only costs about $10.
boozed - Saturday, March 9, 2019 - link
I wonder why it's taken so long