Out of curiosity, has any drive manufacturer explored the possibility of eliminating the actuator movement altogether and instead implementing multiple read/write heads to cover the entire surface area of the disk? I am a bit naive when it comes to hard drive developments. I know how they work, however.
Each track is microns wide, they aren't able to have that many heads. Or rather if they do, things get way bigger way quicker and MUUUUCH more expensive.
It hardly seems inconceivable that some new technology could enable a “single” head covering the entire radius of the platter to address each track individually. I have no idea what that would look like, but obviously trying to cram several million traditional heads into a 3.5” drive is a non-starter.
Or maybe do it with several heads like you said, spreaded on the entire radius of the drive and these heads to have a capability to move a few degrees to cover a number of sectors with each head.
Current hard drives are up to half a million tracks per inch. One head per track per platter is not at all practical. Might not even be geometrically possible to fit that many heads arrayed over the surface of the platter.
The risks of any major new R&D with HDD are too high. You would be committing many billions in hopes of a breakthrough that may or may not come. WD and Seagate have one chance right now to re-invent themselves as a next gen storage company or to remain stuck on hard drives and great risk of liquidation. That is if its not already too late.
Risk of liquidation?? HDDs still have a long long long way to go before they are obsolete. SSDs have no way of matching HDD's cost per GB...So more and more servers are using hybrids RAIDs now. SSD for cache and HDD for storage. If you using only SSD the cost will be much much higher.
" HDDs still have a long long long way to go before they are obsolete. SSDs have no way of matching HDD's cost per GB... "
not after amortizing billions of dollars of R&D over 50 million units shipped per year to the "stronk enterprise stronk demand stronk market" divided between WD, Seagate, toshiba or hitachi or whatever others you have.
Another is for backup. Due to low cost, HDD are used for backup as well. Under such environment, speed is not that critical. Main thing is space. Imagine 100TB of HDD space vs 100TB of SSD for backup....
Another is virtually unlimited writes.... SSDs have problems with lifespan (limited no. of writes). No such issue with HDDs.
The lifespan of the SSD is a non issue, they have such a huge number of writes that it would take a big effort to reach their EOL before they become obsolete or not really worth to keep around. HDD have moving parts so they are in theory more prone to fail.
One of the stronger arguments for magnetic storage is durability. Try leaving both a QLC SSD and a HDD on a shelf, for a couple years, and then we'll talk about HDDs' obsolescence.
A company called Dataslide made waves over a decade ago talking about their "Hard Rectangular Disk" tech that used rectangular platters coupled to some sort of matching size components with arrays of read/write heads - actuation of the platter was piezoelectric in a back and forth motion, rather than the spinning motion of the HDD.
IBM came up with a MEMS "caterpillar" system that had a high density of write heads for a non-rotational heat-assisted "memory" storage tech.
All HDDs made today have a minimum of 2 heads per track - one on either side of the platter.
It would be feasible to put 8 parallel heads per platter side and operate them as a shingle, which would allow them to rewrite the track in a single rotation. That won't be cost-effective today but it may be in the future when we have better nano-scale manufacturing facilities. The challenge isn't so much stacking the heads as it is getting all the same advanced writing features - pmr, eamr, hamr, mamr, working close together without interfering and with minimal duplication of parts, and also the increased risk of failure having 8 heads per platter side instead of one meaning 8x as many things can go wrong. They'd also have to solve the particle ejection problem for the parallel heads - when there's a particle with a single head it will tend to get out of the way. With 2 or more heads there's an increasing risk it gets confined into a collision with the head. Much more than 8 heads per platter side will never be feasible.
The main problem is that the drive platters change size during operation due to thermal expansion. So current drives can't read and write by cylinder, as the relative position of data on different platters is not constant, they have to seek every track.
Your suggestion goes back to the original hard drives. Before using the disk shape, there were drums which had fixed read/write heads. Following that, the original disks also used fixed read/write heads. As heads got closer to the surface, they needed to fly to deal with minute imperfections and thermal deformations in the disk which were not completely flat. The aerodynamics of fixed heads do not allow them to fly independently over the surface of the drive, which would require them to be further from the disk, reducing density.
done by IBM mainframes, sort of, in the 60s. their DASD (as they called them) had multiple (4 or so?) radial actuator sets, each I/O-ing on the entire disc pack. for some time now, even mainframes use conventional 'PC' full-size drives, although more robust and not generally available to civilians.
"Out of curiosity, has any drive manufacturer explored the possibility of eliminating the actuator movement altogether and instead implementing multiple read/write heads to cover the entire surface area of the disk? "
another 'innovation' from the 70s, sort of (thanks to the wiki): "The "x2F", as in Model A2F, unit is a normal x2 unit, but its two HDAs also have a Fixed Head area over the first five cylinders, thereby reducing[c] seek time to zero for these five cylinders. This fixed head area is intended to be allocated to the frequently accessed HASP or JES2 checkpoint area and thus greatly reduce head motion on the SPOOL device. The fixed head area can also be utilized for TSO swap data (MVT and SVS) and system swap data (MVS) wherein the swap data for SVS and MVS consist of blocks of pages that have been in memory when an address space is selected for swap-out; those pages need not be contiguous and in general do not include pages that have not been modified since their last page-in. This system architecture greatly improves context switches between TSO users or batch regions. "
I understand current tech of the platter->head relationship, the platter spinning causes a laminar flow of air over it, making the head "float" on top. But in the end the head floats because of drag or friction. This small bit of friction seems insignificant, but there is a reason why "green" drives park their heads. So there are two parts increasing the number of heads increases the amount of friction thus total increasing both the required energy to maintain spin, and heat dissipation. second part would be that laminar flow of air on top of the disk. The disk only spins so fast, so the air can only hold up so much. if you were to increase the weight of the total weight of the heads, there might not be enough force to keep the heads off the platters causing a crash. Also if the flow were disrupted and became a turbulent flow, the head might be forced down or flopped around
Each recording head costs around $5, making the combined cost of recording heads the largest line item on a bill of materials for a drive with a lot of heads. Being able to use one head over the whole (very cheap) surface of a disc is what makes mechanical drives cheaper than SSDs. If you had to fab one head per track, you would end up with something that cost way more than an SSD, but had much worse performance.
"... Western Digital is still a few years away from deploying either HAMR or MAMR ..."
so years after pretending Energy Assisted Recording was here and it was real, now they admit it was mostly fake news and lies.
clearly they did not learn from intel's experiences (such as 5g and 10nm) and they will likely continue to market products that don't exist as if that is going to save hard drives from much better product that is getting better and cheaper exponentially every year.
I hate HDDs as much as the next guy but... you're just being silly. The technology isn't fake, in fact if you actually read the article you'd know that they are deploying the first gen energy assisted drives - ePMR. The full HAMR and MAMR drives just aren't ready for primetime yet. This has absolutely nothing to do with Intel's fab woes or anything else. Trying to link them as a "lesson" is irrational at best.
They also don't compete with SSDs AT ALL, nor do they claim to. These drives are aimed at enterprise/server (especially cloud) usage where they need the capacity above all else. That aside, SSDs aren't getting exponentially cheaper or better every year. You might want to look the word exponential up, unless you were just trying to be hyperbolic. If anything the price per bit improvements have been slowing down quite a lot in the past year or two.
The spec sheet for the SN340 says 7k IOPS for 32kB random writes. They really don't want anyone getting the wrong idea about what that drive is and is not suitable for.
I read something a long while back about Seagate--or it might've been WD, can't recall--about HDD developers looking at a sort of internal RAID 0 hardware arrangement--like with 2-4 platters (of whatever size) served by one head per platter but reading and writing in RAID 0 format, whether RAID 0x2, x3, or x4, etc. I would imagine the notion has been shelved as I have heard nothing about it since. Still, though, even at RAID 0 x4 (just using "RAID 0" for concept here) that would look only slightly faster than SATA SSD performance and still be much slower than NVMe (standard--not even talking about RAID 0 NVMe performance.) I guess it never panned out...
That's more or less what multi-actuator drives are. Most drives have one head on each side of each platter, but the heads all move as a group, and only one head can be actively reading or writing at a time. A dual actuator drive splits the heads into two groups that can move independently. It basically turns an 8-platter drive into two 4-platter drives that happen to share a spindle motor.
Billy - but what if all heads could read (or write) at the same time while all moving as a group? For an 8 platter drive, that’s 16 heads. For a 4KB minimum block size read, each head would read 256 bytes at the same time, then the drive reassembles that into a 4KB block to send to the OS. Same for writing.
Difficult, with a number of tricky issues, but inherently not impossible especially with how crazy HDD technology already is with regards to head size and precision.
Unlike Seagate's dual actuator design, this wouldn't improve random performance, but it would give much faster sequential performance, and I don't see why hard disk manufacturers haven't done it. The main technical issue is that you have to be able to move the heads small distances independently of each other to ensure that each head is precisely centered over the track it is reading or writing. So once the disk drive manufacturers started using multi-stage actuators (where each head has its own second stage), allowing the individual heads to read or write in parallel seems like an obvious next step.
Very interesting thank you Kenneth. With SSDs there is no reason for HDDs to improve random access, but the sheer amount of time needed to load or copy off a huge HDD (30-50TB+) is a concern. With this tech a big jump in sequential speed would be possible. Also it could improve reliability as with the right cylinder coding, a HDD could fully recover data from a bad head crash. Could even put one or several heads fully out of commission and safely continue, albeit more slowly.
I hope both SMR for HDDs and "Zoned Storage Initiative" for SSDs go away. Both are terrible performers for little-to-no gain in savings for the customer. Move all of the engineers working on it to the energy-assisted technology departments, and let those poor-performers (especially SMR) fade into obscurity.
"For SSDs, switching to a zoned model can allow for drastically smaller overprovisioning ratios and onboard DRAM, so drives can be cheaper while offering similar performance on many workloads." Translating WD's marketing spin - worse performance (often significantly) while not always offering much in the way of price savings to the customer. I hope this "Zoned Storage Initiative" never makes it.
I don't like SMR any more than you do, but there are sequential use cases (i.e. backups & extremely large files) where it works just fine. And it's still a heck of a lot better than tape!
I do agree when you look at the current situation for SMR HDDs. However, as the article/WD mentions, the gap between CMR and SMR HDDs will widen in the very near future. Let's imagine you can get a 18TB CMR HDD or a 30TB SMR HDD for the same price. I think in that case, people will start asking themselves whether the SMR variant could make sense for certain workloads (as mode_13h mentioned, backups, large files, mostly read-only stuff etc.).
But again - I agree, in the current situation, I steer away from SMR drives because there is no advantage apart from small price savings but huge disadvantages. And I guess pretty much everyone is feeling the same way, if you look at the projections from WD.
Still struck by the "energy assisted". So, what is it? They'll make the magnetic substrate a bit warmer, but not with a laser or microwave? Unless more is known how WD achieves their greater density, many people won't trust their data to these disks. I like to know what technology I am supposed to entrust my data with.
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
39 Comments
Back to Article
eek2121 - Friday, January 31, 2020 - link
Out of curiosity, has any drive manufacturer explored the possibility of eliminating the actuator movement altogether and instead implementing multiple read/write heads to cover the entire surface area of the disk? I am a bit naive when it comes to hard drive developments. I know how they work, however.Drizzt321 - Friday, January 31, 2020 - link
Each track is microns wide, they aren't able to have that many heads. Or rather if they do, things get way bigger way quicker and MUUUUCH more expensive.chaos215bar2 - Sunday, February 2, 2020 - link
It hardly seems inconceivable that some new technology could enable a “single” head covering the entire radius of the platter to address each track individually. I have no idea what that would look like, but obviously trying to cram several million traditional heads into a 3.5” drive is a non-starter.yeeeeman - Monday, February 3, 2020 - link
Or maybe do it with several heads like you said, spreaded on the entire radius of the drive and these heads to have a capability to move a few degrees to cover a number of sectors with each head.Soulkeeper - Monday, February 3, 2020 - link
I've always thought that a "read strip" or "read bar" that spans the radius of the platter might be doable. I guess it's just too expensive ...Billy Tallis - Friday, January 31, 2020 - link
Current hard drives are up to half a million tracks per inch. One head per track per platter is not at all practical. Might not even be geometrically possible to fit that many heads arrayed over the surface of the platter.mode_13h - Saturday, February 8, 2020 - link
I think the idea is to use a phased array. Yet, even if that got you something like a 10:1 reduction in heads per track, it'd still be impractical.azfacea - Friday, January 31, 2020 - link
The risks of any major new R&D with HDD are too high. You would be committing many billions in hopes of a breakthrough that may or may not come. WD and Seagate have one chance right now to re-invent themselves as a next gen storage company or to remain stuck on hard drives and great risk of liquidation. That is if its not already too late.escksu - Sunday, February 2, 2020 - link
Risk of liquidation?? HDDs still have a long long long way to go before they are obsolete. SSDs have no way of matching HDD's cost per GB...So more and more servers are using hybrids RAIDs now. SSD for cache and HDD for storage. If you using only SSD the cost will be much much higher.azfacea - Tuesday, February 4, 2020 - link
" HDDs still have a long long long way to go before they are obsolete. SSDs have no way of matching HDD's cost per GB... "not after amortizing billions of dollars of R&D over 50 million units shipped per year to the "stronk enterprise stronk demand stronk market" divided between WD, Seagate, toshiba or hitachi or whatever others you have.
azfacea - Tuesday, February 4, 2020 - link
but hey if you said long a few more times maybe you become rightescksu - Sunday, February 2, 2020 - link
Another is for backup. Due to low cost, HDD are used for backup as well. Under such environment, speed is not that critical. Main thing is space. Imagine 100TB of HDD space vs 100TB of SSD for backup....Another is virtually unlimited writes.... SSDs have problems with lifespan (limited no. of writes). No such issue with HDDs.
Strunf - Monday, February 3, 2020 - link
The lifespan of the SSD is a non issue, they have such a huge number of writes that it would take a big effort to reach their EOL before they become obsolete or not really worth to keep around. HDD have moving parts so they are in theory more prone to fail.mode_13h - Saturday, February 8, 2020 - link
One of the stronger arguments for magnetic storage is durability. Try leaving both a QLC SSD and a HDD on a shelf, for a couple years, and then we'll talk about HDDs' obsolescence.soresu - Saturday, February 1, 2020 - link
A company called Dataslide made waves over a decade ago talking about their "Hard Rectangular Disk" tech that used rectangular platters coupled to some sort of matching size components with arrays of read/write heads - actuation of the platter was piezoelectric in a back and forth motion, rather than the spinning motion of the HDD.mode_13h - Saturday, February 8, 2020 - link
Man, what a bad idea. So many problems with moving an array of heads in a linear fashion... no wonder it failed.linuxgeex - Sunday, February 2, 2020 - link
IBM came up with a MEMS "caterpillar" system that had a high density of write heads for a non-rotational heat-assisted "memory" storage tech.All HDDs made today have a minimum of 2 heads per track - one on either side of the platter.
It would be feasible to put 8 parallel heads per platter side and operate them as a shingle, which would allow them to rewrite the track in a single rotation. That won't be cost-effective today but it may be in the future when we have better nano-scale manufacturing facilities. The challenge isn't so much stacking the heads as it is getting all the same advanced writing features - pmr, eamr, hamr, mamr, working close together without interfering and with minimal duplication of parts, and also the increased risk of failure having 8 heads per platter side instead of one meaning 8x as many things can go wrong. They'd also have to solve the particle ejection problem for the parallel heads - when there's a particle with a single head it will tend to get out of the way. With 2 or more heads there's an increasing risk it gets confined into a collision with the head. Much more than 8 heads per platter side will never be feasible.
quadibloc - Monday, February 3, 2020 - link
The main problem is that the drive platters change size during operation due to thermal expansion. So current drives can't read and write by cylinder, as the relative position of data on different platters is not constant, they have to seek every track.jhh - Monday, February 3, 2020 - link
Your suggestion goes back to the original hard drives. Before using the disk shape, there were drums which had fixed read/write heads. Following that, the original disks also used fixed read/write heads. As heads got closer to the surface, they needed to fly to deal with minute imperfections and thermal deformations in the disk which were not completely flat. The aerodynamics of fixed heads do not allow them to fly independently over the surface of the drive, which would require them to be further from the disk, reducing density.FunBunny2 - Monday, February 3, 2020 - link
done by IBM mainframes, sort of, in the 60s. their DASD (as they called them) had multiple (4 or so?) radial actuator sets, each I/O-ing on the entire disc pack. for some time now, even mainframes use conventional 'PC' full-size drives, although more robust and not generally available to civilians.FunBunny2 - Monday, February 3, 2020 - link
"Out of curiosity, has any drive manufacturer explored the possibility of eliminating the actuator movement altogether and instead implementing multiple read/write heads to cover the entire surface area of the disk? "another 'innovation' from the 70s, sort of (thanks to the wiki): "The "x2F", as in Model A2F, unit is a normal x2 unit, but its two HDAs also have a Fixed Head area over the first five cylinders, thereby reducing[c] seek time to zero for these five cylinders. This fixed head area is intended to be allocated to the frequently accessed HASP or JES2 checkpoint area and thus greatly reduce head motion on the SPOOL device. The fixed head area can also be utilized for TSO swap data (MVT and SVS) and system swap data (MVS) wherein the swap data for SVS and MVS consist of blocks of pages that have been in memory when an address space is selected for swap-out; those pages need not be contiguous and in general do not include pages that have not been modified since their last page-in. This system architecture greatly improves context switches between TSO users or batch regions. "
bebimbap - Tuesday, February 4, 2020 - link
I understand current tech of the platter->head relationship, the platter spinning causes a laminar flow of air over it, making the head "float" on top. But in the end the head floats because of drag or friction. This small bit of friction seems insignificant, but there is a reason why "green" drives park their heads.So there are two parts increasing the number of heads increases the amount of friction thus total increasing both the required energy to maintain spin, and heat dissipation.
second part would be that laminar flow of air on top of the disk. The disk only spins so fast, so the air can only hold up so much. if you were to increase the weight of the total weight of the heads, there might not be enough force to keep the heads off the platters causing a crash. Also if the flow were disrupted and became a turbulent flow, the head might be forced down or flopped around
0x4C4B - Monday, February 10, 2020 - link
Each recording head costs around $5, making the combined cost of recording heads the largest line item on a bill of materials for a drive with a lot of heads. Being able to use one head over the whole (very cheap) surface of a disc is what makes mechanical drives cheaper than SSDs. If you had to fab one head per track, you would end up with something that cost way more than an SSD, but had much worse performance.azfacea - Friday, January 31, 2020 - link
"... Western Digital is still a few years away from deploying either HAMR or MAMR ..."so years after pretending Energy Assisted Recording was here and it was real, now they admit it was mostly fake news and lies.
clearly they did not learn from intel's experiences (such as 5g and 10nm) and they will likely continue to market products that don't exist as if that is going to save hard drives from much better product that is getting better and cheaper exponentially every year.
Alexvrb - Friday, January 31, 2020 - link
I hate HDDs as much as the next guy but... you're just being silly. The technology isn't fake, in fact if you actually read the article you'd know that they are deploying the first gen energy assisted drives - ePMR. The full HAMR and MAMR drives just aren't ready for primetime yet. This has absolutely nothing to do with Intel's fab woes or anything else. Trying to link them as a "lesson" is irrational at best.They also don't compete with SSDs AT ALL, nor do they claim to. These drives are aimed at enterprise/server (especially cloud) usage where they need the capacity above all else. That aside, SSDs aren't getting exponentially cheaper or better every year. You might want to look the word exponential up, unless you were just trying to be hyperbolic. If anything the price per bit improvements have been slowing down quite a lot in the past year or two.
RedGreenBlue - Friday, January 31, 2020 - link
“this means the SN340 will allow the host to issue random write commands, but performance for those will suck.”Haha... Tell us how you really feel.
Billy Tallis - Friday, January 31, 2020 - link
The spec sheet for the SN340 says 7k IOPS for 32kB random writes. They really don't want anyone getting the wrong idea about what that drive is and is not suitable for.ksec - Saturday, February 1, 2020 - link
This is sad as the technology keeps getting pushed back. Although I know I am the minority in terms of HDD usage.WaltC - Saturday, February 1, 2020 - link
I read something a long while back about Seagate--or it might've been WD, can't recall--about HDD developers looking at a sort of internal RAID 0 hardware arrangement--like with 2-4 platters (of whatever size) served by one head per platter but reading and writing in RAID 0 format, whether RAID 0x2, x3, or x4, etc. I would imagine the notion has been shelved as I have heard nothing about it since. Still, though, even at RAID 0 x4 (just using "RAID 0" for concept here) that would look only slightly faster than SATA SSD performance and still be much slower than NVMe (standard--not even talking about RAID 0 NVMe performance.) I guess it never panned out...Billy Tallis - Saturday, February 1, 2020 - link
That's more or less what multi-actuator drives are. Most drives have one head on each side of each platter, but the heads all move as a group, and only one head can be actively reading or writing at a time. A dual actuator drive splits the heads into two groups that can move independently. It basically turns an 8-platter drive into two 4-platter drives that happen to share a spindle motor.Tomatotech - Sunday, February 2, 2020 - link
Billy - but what if all heads could read (or write) at the same time while all moving as a group? For an 8 platter drive, that’s 16 heads. For a 4KB minimum block size read, each head would read 256 bytes at the same time, then the drive reassembles that into a 4KB block to send to the OS. Same for writing.Difficult, with a number of tricky issues, but inherently not impossible especially with how crazy HDD technology already is with regards to head size and precision.
KennethAlmquist - Sunday, February 2, 2020 - link
Unlike Seagate's dual actuator design, this wouldn't improve random performance, but it would give much faster sequential performance, and I don't see why hard disk manufacturers haven't done it. The main technical issue is that you have to be able to move the heads small distances independently of each other to ensure that each head is precisely centered over the track it is reading or writing. So once the disk drive manufacturers started using multi-stage actuators (where each head has its own second stage), allowing the individual heads to read or write in parallel seems like an obvious next step.Tomatotech - Monday, February 3, 2020 - link
Very interesting thank you Kenneth. With SSDs there is no reason for HDDs to improve random access, but the sheer amount of time needed to load or copy off a huge HDD (30-50TB+) is a concern. With this tech a big jump in sequential speed would be possible. Also it could improve reliability as with the right cylinder coding, a HDD could fully recover data from a bad head crash. Could even put one or several heads fully out of commission and safely continue, albeit more slowly.WaltC - Sunday, February 2, 2020 - link
Ah-ha...Thank you for the explanation!romrunning - Monday, February 3, 2020 - link
I hope both SMR for HDDs and "Zoned Storage Initiative" for SSDs go away. Both are terrible performers for little-to-no gain in savings for the customer. Move all of the engineers working on it to the energy-assisted technology departments, and let those poor-performers (especially SMR) fade into obscurity."For SSDs, switching to a zoned model can allow for drastically smaller overprovisioning ratios and onboard DRAM, so drives can be cheaper while offering similar performance on many workloads." Translating WD's marketing spin - worse performance (often significantly) while not always offering much in the way of price savings to the customer. I hope this "Zoned Storage Initiative" never makes it.
mode_13h - Saturday, February 8, 2020 - link
I don't like SMR any more than you do, but there are sequential use cases (i.e. backups & extremely large files) where it works just fine. And it's still a heck of a lot better than tape!Chloiber - Sunday, February 9, 2020 - link
I do agree when you look at the current situation for SMR HDDs. However, as the article/WD mentions, the gap between CMR and SMR HDDs will widen in the very near future. Let's imagine you can get a 18TB CMR HDD or a 30TB SMR HDD for the same price. I think in that case, people will start asking themselves whether the SMR variant could make sense for certain workloads (as mode_13h mentioned, backups, large files, mostly read-only stuff etc.).But again - I agree, in the current situation, I steer away from SMR drives because there is no advantage apart from small price savings but huge disadvantages. And I guess pretty much everyone is feeling the same way, if you look at the projections from WD.
eastcoast_pete - Tuesday, February 4, 2020 - link
Still struck by the "energy assisted". So, what is it? They'll make the magnetic substrate a bit warmer, but not with a laser or microwave? Unless more is known how WD achieves their greater density, many people won't trust their data to these disks. I like to know what technology I am supposed to entrust my data with.Qasar - Wednesday, February 5, 2020 - link
zip discs or optical :-)