0 = Instant Money Support 4 = Like so, we flipped a bit on 0 to 1, and like now it takes a few days for what flippin' that bit did. Meaning, like you pay us, like a ~lil extra~ and we'll like, flip that bit back to 0 for ya. Ain't we patron saints?
These massive capacity drives tend to be rather expensive, and a single pass overwrite can take > 24 hours. That makes instant secure erase an attractive feature if you deal with customer / confidential / regulated data and want to be able to quickly repurpose a drive instead of just shredding it and shelling out for a new one.
Instant secure erase is enabled by making the drive self encrypting and just chucking the keys when you issue a SATA secure erase command. However, if anything happens to the controller where those keys are stored you end up with a failed drive and very little possibility of performing data recovery. If the drive is being used for archival of non-sensitive data, customers might prefer to fail-safe rather than fail-secure.
Could HGST have provided a single SKU along with a firmware tool that allows customers to choose which mode they want? Maybe. But seeing as certain regions also impose regulations on self encrypting drive technology, they may have decided it was easier to just offer two different models.
(I also have reading comprehension fail, as you can tell from reading my response which came from a misreading of your very clever question about Dracala. Reading comprehension fails happen a lot in Zamunda.)
I used to work for a large company that’s very much in the target market for SMR. Our guys looked at it and made it work but ultimately you didn’t get enough from SMR vs either waiting slightly longer or accepting a bit less capacity with a conventional drive.
I don't think the management requirement comes from the drive being SMR, it's because it's a "Host-Managed" drive.
Which to me, "Host-Managed" = we couldn't figure out a way to have the integrated controller be performant across a wide range of usage scenarios, so we're pawning that off to the OS folks to treat this as glorified tape storage.
I know that sounds mean, and I don't mean to downplay the engineers who made this drive possible. The fact that "Host-Managed" is even a thing honestly shows just how super hard it would be to make it "all rounder" performant - it might not even be possible.
SMR really is the sacrifice of random I/O performance for capacity - which makes total sense, since these days if you need read/write perf you use solid state.
These drives could be the drives that offer more backup for cloud servers. For sure there a ton of applications for these drives. The future for speed is for sure SSD (that are fast enough these days) but capacity will be magnetic. Why not mix those ?
xrror gave the gist of this. It's possible to make an SMR drive that works well for certain use-cases and just-about-tolerably for others, but it won't be optimal for any of them - and in some cases (notably sustained random writes) the performance degrades catastrophically.
There's been a fair amount of research on SMR over the past decade which analyzed the approaches and how they've been taken - not always successfully by the manufacturers.
The OS or software, in theory, use its knowledge of the situation and the disk layout and capabilities to better-determine the ideal physical placement of the data and metadata being stored.
To take a simple example: if you *know* you'll only ever write forwards - which might well be a use-case for "archiving" situations, e.g. Facebook content - you could theoretically provide maximum efficiency by simply having a single set of shingles, without any of the normal breaks that allow you to delete old data and reuse the space.
You might alternatively create one shingle per variably-sized (but large) data blob - or use fixed sizes to decrease fragmentation, similar to a buddy memory allocator: https://en.wikipedia.org/wiki/Buddy_memory_allocat...
This is the world of software-defined disk drives. It means hard drives become an increasingly specialized, bespoke form of storage - rather like tape.
This kind of software/firmware mix happens in other areas, too, like CPU core boosting. Either the OS can do it or more recently it tends to tell tells the CPU to handle it and gives it hints about priority (processor core control). One or the other can be more optimal depending on the circumstances.
My first experience with SMR was the Apple Disk][ Shugart drive. The write energy resulted in a track twice as wide as the head needed to accurately read it back. It was possible to step the drive by quarter tracks, and if you were careful with the track positioning - always step up past where you want to go then step down - then it was possible to write overlapping half tracks and read them back just fine. The problem came when you wanted to write to them again because every 1/2 overlapping track would get obliterated, because the read head was effectively over both at the same time. So you could squeeze 7 tracks into 4 tracks of space. If you dropped the gap between sectors and used the sector Prologue as the Epilogue as well, then it was possible to fit 22 sectors into a track instead of 16. Between the two you could fit slightly over double the total storage on a floppy. But there was a problem. You needed to hold all 7 tracks in memory at once, plus an MFM-encoded nybble buffer for one of the 7 tracks, which took basically twice the space once you also fit the prologues. So you needed 8x22x256 bytes = 45k of RAM to be able to write a single sector to the disk and that took about 63 revolutions at 360RPM, or about 200ms to read it and about the same to write it back. DOS required about 8k out of the system's 48k, about 4k was used by the OS and display hardware ($0000-$077) so it wasn't possible to do this without a 16k Language Card. So although this was great for archiving read-only stuff, it was way too inconvenient to catch on. But read-only you could have a floppy with double the data on it and surprise your friends by packing more games on a disk, and none of the mainstream disk copy programs could copy them... not even the bit-nybblers like Locksmith, because they didn't know how to rewrite the Epilogues.
Windows does not recognize SMR disks as of now. That being said, if you know your having a SMR drive, you can easily mitigate the drive into eg. backup rutines. SMR is great for streamed writes, but the more random or deletes, the more they will be a bad choice. SMR is neither good for raid5+. The 512MB cache is to try to avoid too much of these random writes hits the disk. That being said, i had my fair experience with SMR disks and they work fine for my usage (backup). But, as others already wrote, then the SMR disks ain't worth it, vs the price they offer. SMR was promised in initial postings, that it would give 20% more space for same value. There already exists 14TB none SMR disks in the market, so we should see almost 17TB theoretical with SMR disks, OR we should see 14TB SMR disks with 20% less price. That's not what were seeing now..
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.
19 Comments
Back to Article
olafgarten - Friday, October 26, 2018 - link
What is the difference between the M0 and M4 models?repoman27 - Friday, October 26, 2018 - link
M = Host-Managed0 = Instant Secure Erase support
4 = Secure Erase support (overwrite only)
alacard - Friday, October 26, 2018 - link
0 = Instant Money Support4 = Like so, we flipped a bit on 0 to 1, and like now it takes a few days for what flippin' that bit did.
Meaning, like you pay us, like a ~lil extra~ and we'll like, flip that bit back to 0 for ya.
Ain't we patron saints?
Lord of the Bored - Saturday, October 27, 2018 - link
I really do hate how the modern computer component market works.repoman27 - Sunday, October 28, 2018 - link
These massive capacity drives tend to be rather expensive, and a single pass overwrite can take > 24 hours. That makes instant secure erase an attractive feature if you deal with customer / confidential / regulated data and want to be able to quickly repurpose a drive instead of just shredding it and shelling out for a new one.Instant secure erase is enabled by making the drive self encrypting and just chucking the keys when you issue a SATA secure erase command. However, if anything happens to the controller where those keys are stored you end up with a failed drive and very little possibility of performing data recovery. If the drive is being used for archival of non-sensitive data, customers might prefer to fail-safe rather than fail-secure.
Could HGST have provided a single SKU along with a firmware tool that allows customers to choose which mode they want? Maybe. But seeing as certain regions also impose regulations on self encrypting drive technology, they may have decided it was easier to just offer two different models.
alacard - Sunday, October 28, 2018 - link
That's very interesting. It's always nice to learn something new, thanks.Manch - Tuesday, October 30, 2018 - link
Are you the son of Dracala?! :Dalacard - Tuesday, October 30, 2018 - link
I think you have me confused with Alucard.I'm alacard, son of dracala (a goat herder from Zamunda).
alacard - Tuesday, October 30, 2018 - link
(I also have reading comprehension fail, as you can tell from reading my response which came from a misreading of your very clever question about Dracala. Reading comprehension fails happen a lot in Zamunda.)Manch - Friday, November 2, 2018 - link
LOL, awesome!Tchamber - Friday, October 26, 2018 - link
"The manufacturer will only sell these products to customers with software that can manage SMR hard drives,"clearly, these aren't consumer HDDs, but what software does an SMR HDD require for management? Does Windows not recognize SMR?
allenb - Friday, October 26, 2018 - link
I used to work for a large company that’s very much in the target market for SMR. Our guys looked at it and made it work but ultimately you didn’t get enough from SMR vs either waiting slightly longer or accepting a bit less capacity with a conventional drive.vFunct - Saturday, October 27, 2018 - link
Seems like SMR drives should really be 3x or more denser for all the custom work you need to do to make them work.xrror - Friday, October 26, 2018 - link
I don't think the management requirement comes from the drive being SMR, it's because it's a "Host-Managed" drive.Which to me, "Host-Managed" = we couldn't figure out a way to have the integrated controller be performant across a wide range of usage scenarios, so we're pawning that off to the OS folks to treat this as glorified tape storage.
I know that sounds mean, and I don't mean to downplay the engineers who made this drive possible. The fact that "Host-Managed" is even a thing honestly shows just how super hard it would be to make it "all rounder" performant - it might not even be possible.
SMR really is the sacrifice of random I/O performance for capacity - which makes total sense, since these days if you need read/write perf you use solid state.
RaduR - Saturday, October 27, 2018 - link
These drives could be the drives that offer more backup for cloud servers. For sure there a ton of applications for these drives.The future for speed is for sure SSD (that are fast enough these days) but capacity will be magnetic. Why not mix those ?
GreenReaper - Saturday, October 27, 2018 - link
xrror gave the gist of this. It's possible to make an SMR drive that works well for certain use-cases and just-about-tolerably for others, but it won't be optimal for any of them - and in some cases (notably sustained random writes) the performance degrades catastrophically.There's been a fair amount of research on SMR over the past decade which analyzed the approaches and how they've been taken - not always successfully by the manufacturers.
The OS or software, in theory, use its knowledge of the situation and the disk layout and capabilities to better-determine the ideal physical placement of the data and metadata being stored.
To take a simple example: if you *know* you'll only ever write forwards - which might well be a use-case for "archiving" situations, e.g. Facebook content - you could theoretically provide maximum efficiency by simply having a single set of shingles, without any of the normal breaks that allow you to delete old data and reuse the space.
You might alternatively create one shingle per variably-sized (but large) data blob - or use fixed sizes to decrease fragmentation, similar to a buddy memory allocator:
https://en.wikipedia.org/wiki/Buddy_memory_allocat...
This is the world of software-defined disk drives. It means hard drives become an increasingly specialized, bespoke form of storage - rather like tape.
This kind of software/firmware mix happens in other areas, too, like CPU core boosting. Either the OS can do it or more recently it tends to tell tells the CPU to handle it and gives it hints about priority (processor core control). One or the other can be more optimal depending on the circumstances.
linuxgeex - Sunday, October 28, 2018 - link
My first experience with SMR was the Apple Disk][ Shugart drive. The write energy resulted in a track twice as wide as the head needed to accurately read it back. It was possible to step the drive by quarter tracks, and if you were careful with the track positioning - always step up past where you want to go then step down - then it was possible to write overlapping half tracks and read them back just fine. The problem came when you wanted to write to them again because every 1/2 overlapping track would get obliterated, because the read head was effectively over both at the same time. So you could squeeze 7 tracks into 4 tracks of space. If you dropped the gap between sectors and used the sector Prologue as the Epilogue as well, then it was possible to fit 22 sectors into a track instead of 16. Between the two you could fit slightly over double the total storage on a floppy. But there was a problem. You needed to hold all 7 tracks in memory at once, plus an MFM-encoded nybble buffer for one of the 7 tracks, which took basically twice the space once you also fit the prologues. So you needed 8x22x256 bytes = 45k of RAM to be able to write a single sector to the disk and that took about 63 revolutions at 360RPM, or about 200ms to read it and about the same to write it back. DOS required about 8k out of the system's 48k, about 4k was used by the OS and display hardware ($0000-$077) so it wasn't possible to do this without a 16k Language Card. So although this was great for archiving read-only stuff, it was way too inconvenient to catch on. But read-only you could have a floppy with double the data on it and surprise your friends by packing more games on a disk, and none of the mainstream disk copy programs could copy them... not even the bit-nybblers like Locksmith, because they didn't know how to rewrite the Epilogues.marras - Saturday, October 27, 2018 - link
Windows does not recognize SMR disks as of now.That being said, if you know your having a SMR drive, you can easily mitigate the drive into eg. backup rutines. SMR is great for streamed writes, but the more random or deletes, the more they will be a bad choice. SMR is neither good for raid5+. The 512MB cache is to try to avoid too much of these random writes hits the disk.
That being said, i had my fair experience with SMR disks and they work fine for my usage (backup).
But, as others already wrote, then the SMR disks ain't worth it, vs the price they offer.
SMR was promised in initial postings, that it would give 20% more space for same value.
There already exists 14TB none SMR disks in the market, so we should see almost 17TB theoretical with SMR disks, OR we should see 14TB SMR disks with 20% less price.
That's not what were seeing now..
VenkataSwami - Wednesday, September 9, 2020 - link
WD HM-SMR drives, what and all File systems it is supporting? Is it supports XFS, EXT4 File systems?