SoylentNews is people
SoylentNews
Western Digital Announces 22TB CMR and 26TB SMR HDDs: 10 Platters plus ePMR
Western Digital is announcing the sampling of its new 22TB CMR and 26TB SMR hard drives today at its What's Next Western Digital Event. As usual, the hyperscale cloud customers will get first dibs on these drives. The key takeaway from today's presentation is that Western Digital doesn't yet feel the need to bring heat-assisted magnetic recording (HAMR) into the picture. In fact, WD is doubling down on energy-assisted PMR (ePMR) technology and OptiNAND (introduced first in the 20TB CMR drives). WD is also continuing to use the triple-stage actuator that it started shipping in the first half of 2020 in the new drives. It goes without saying that the new high-capacity drives are helium-filled (HelioSeal technology). The main change common to both drives is the shift to a 10-stack design.
The SMR drives are getting an added capacity boost, thanks to WD's new UltraSMR technology. This involves adoption of a new advanced error correction algorithm to go along with encoding of larger blocks. This allows improvement in the tracks-per-inch (TPI) metric, resulting in 2.6TB per platter. The new Ultrastar DC HC670 uses ten platters to provide 26TB of host-managed SMR storage for cloud service providers.
PMR = Perpendicular Magnetic Recording
SMR = Shingled Magnetic Recording
OptiNAND = embedded flash drive included on the HDD for caching metadata
While the company did not quantify the amount of NAND in its OptiNAND drives, they are stressing the fact that it is not a hybrid drive (SSHD). Unlike SSHDs, the OptiNAND drives do not store any user data at all during normal operation. Instead, the NAND is being used to store metadata from HDD operation in order to improve capacity, performance, and reliability.
(Score: 2) by takyon on Wednesday May 11 2022, @01:30PM (23 children)
Mixed shingled/non-shingled sounds complex. They should just increase the size of the DRAM and NAND caches as necessary. If they need to add 1 TB of NAND, they'll just do that.
OptiNAND does not store data, but surprisingly there are still 3 companies making HDDs and they could take different approaches.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 0) by Anonymous Coward on Wednesday May 11 2022, @01:54PM (2 children)
eh? why not use two drives PMR and SMR in raid config that doesn't have a name (yet)?
so write to PMR, your "stuff" then when smartdrv.sys notices that PMR disk is idyle, start busying it and slowly (that is max speed the SMR will except incoming data) moving data to the SMR... then freeing that data from the PMR disk ... and in the process waiting for a kb to fix smartdrv.sys 'cause it keeps forgetting to update the new location of the data in the filesystem "database" :D
(Score: 2) by RS3 on Wednesday May 11 2022, @04:48PM (1 child)
You're being semi-serious, and I get it. I agree that the best system would involve a much larger cache which would be a combination of RAM and PMR (CMR) and/or SSD and a very intelligent controller to manage it all.
There are hybrid drives. In fact, an SMR drive writes data in PMR initially. The slowdown happens when it must go back and rewrite those tracks in SMR. The problem is: if you write SMR adjacent to a PMR track, you'll damage or destroy the PMR data, so you have to read that track and rewrite it in SMR. Obviously as a drive fills with data scattered everywhere on the disk, you have a complex and speed-crippling situation of doing lots of rewrites.
The Wiki page [wikipedia.org] describes the 3 types of management: drive managed, host managed, and host aware. They also list some Linux filesystems that are "zone aware" and work well with host managed drives (and I have no idea which drives can be host managed).
(Score: 0) by Anonymous Coward on Wednesday May 11 2022, @07:54PM
"best system would involve a much larger cache which would be a combination of RAM and PMR (CMR) and/or SSD and a very intelligent controller to manage it all."
Yes, it's called a computer.
(Score: 2, Interesting) by pTamok on Wednesday May 11 2022, @02:35PM (19 children)
It shouldn't be. If you make one platter non-shingled, then you just need a non-shingled head assembly for that platter. You can then fill up that platter at full-speed write rates, and during idle time migrate the data to the shingled backing store on the other platters.
Even if you use the same head assembly for each platter, if you make the track spacing wider on part of a platter so that shingling is not necessary, you can do faster writes - it's 'just' a question of positioning the head arm slightly differently. You could reserve the inner tracks for this, for example, to minimize seek times during write. I've no idea what the relative cost of doing this compared to using a NAND cache would be, but it's a trade-off between maximising capacity and spending money on a RAM/NAND cache. It's basically a non-volatile on-disk cache implemented in a different way. There might be a niche for it, or not.
(Score: 0) by Anonymous Coward on Wednesday May 11 2022, @02:49PM (10 children)
So you take an SMR drive's slow performance and make the drive even slower by requiring 2 writes instead of one: the first to the non-SMR part of the drive, and the second to the SMR part of the drive. And for this to even have a chance of working, you need long periods of inactivity from the user for I/O request so the drive can do its writing to the non-SMR part of the drive. If the user isn't going to be hitting the drive hard anyway, you might as well just go with a 100% SMR drive and pay the price of only a single drive write instead of two.
I think you will find that a hybrid drive technology is only practical with separate physical drives of each technology where each drive handles a filesystem partition with an expected workload that is suited to that drive technology. Then slow writes don't block fast random I/O.
(Score: 0) by Anonymous Coward on Wednesday May 11 2022, @03:36PM
yes yes. two writes. it only would make sense if SMR are half the price of a PMR (same capacity) and/or PMR cannot be made at same capacity without increasing form factor (and you are thus size constraint for whatever reason).
(Score: 3, Informative) by RS3 on Wednesday May 11 2022, @04:53PM (8 children)
2 writes is literally how SMR drives work, and why they're so slow.
https://en.wikipedia.org/wiki/Shingled_magnetic_recording [wikipedia.org]
Okay, to be fair, sometimes you'll get to write data SMR immediately. But as a drive fills, with data scattered everywhere, you're going to do more and more rewrites, and the moves will be farther and farther away requiring much head movement and settle time overhead.
(Score: 0) by Anonymous Coward on Wednesday May 11 2022, @05:26PM (7 children)
I did not know that SMR was such a stupid design. Who the hell would want such a thing? Besides the slowness, it has double the wear and tear for writes AT LEAST and probably more as it is likely the disk will be fragmented from the double write alogorithm.
(Score: 3, Interesting) by RS3 on Wednesday May 11 2022, @10:08PM (6 children)
Now you're catching on. Well, it increases storage capacity by 20% or so, as you can see in TFS's linked article (22 TB for PMR vs 26 TB for SMR).
I think ultimately it reduces fragmentation because it's always trying to rewrite everything in SMR mode, which I think it has to do in one direction, so eventually you'll still have file fragmentation, but no unused spaces interspersed with the data.
SMR also increases the sales of backup media, software, and systems because your data is, possibly significantly, less safe on SMR.
Point is: drive manufacturers must make this known, and I'm sure there are already many laws regarding truth in advertising, truth in product description and specifications, etc.
(Score: 2) by Reziac on Thursday May 12 2022, @02:55AM (5 children)
And my next question was going to be...
How the hell do you back up a 26TB hard drive?
Yeah, yeah, with another 26TB hard drive... or several of them... or a dozen smaller older ones that you know are reliable...
And there is no Alkibiades to come back and save us from ourselves.
(Score: 0) by Anonymous Coward on Thursday May 12 2022, @06:07AM
We would back up a 26TB hard drive with 6 tapes, 6 Blu-rays, and cold storage fees.
(Score: 3, Interesting) by RS3 on Thursday May 12 2022, @06:09AM (3 children)
Well, 80-column punched cards [wikipedia.org] are inefficient. 96-column cards [wikipedia.org] pack much more data into a much smaller card, so that choice is obvious.
You could use paper tape. It's quite reliable.
While that's keeping you busy changing spools ;-} , I'll use some of these (hang on to your hat if you don't know about this stuff!) :
https://interestingengineering.com/new-magnetic-tape-delivers-a-record-580tb-storage-capacity#:~:text=The%20new%20tape%20produced%20by,inch%20%E2%80%94%206.45%20sq%20cm). [interestingengineering.com]
And even these guys:
https://www.quantum.com/en/products/tape-storage/lto-tape-drives/ [quantum.com]
(Score: 3, Interesting) by Reziac on Thursday May 12 2022, @06:20AM
Holy crap, tape libraries have finally gotten ahead of disks!
Now, where is my wealthy donor so I can buy one of these marvels??
My high school had an IBM1620 that was used for classwork (Fortran II, no less). Punch cards, I know them well... it was a Big Deal when we got a paper tape reader for loading the OS. So much faster -- five minutes instead of half a hour!!
And there is no Alkibiades to come back and save us from ourselves.
(Score: 2) by janrinok on Thursday May 12 2022, @12:39PM (1 child)
(Score: 3, Informative) by RS3 on Thursday May 12 2022, @06:09PM
ebay, sometimes craigslist, etc. Look for LTO. Business depreciates perfectly good equipment. Sadly too often ends up scrapped or in landfill, but some smart people sell it.
Often listers don't know what they have, and just list the drive by model number, rather than "LTO" "drive" "320 GB", or whatever the native capacity is. So some research might be needed, but that's easy, and you might get a better bargain in those cases (I have). Dell sells a lot of drives, but someone else made them of course.
You're probably aware, but in case you're not: be aware that tapes and drives are often labeled / advertised for their compressed capacity. You don't have to use drive's compression- I never trust it. Look for native capacity, which is usually considered 1/2 of the compressed capacity.
I had a longer post going but no time to finish it. But consider there are cheap small tape libraries available too. I have one that holds, iirc, 8 DAT tapes. Another that holds iirc 7 8 mm tapes (yes, like the 8 mm video tape, but qualified for data). So you can get a lot of backup capacity with lots of cheap tapes and not have to manually change tapes as often (if your backup spans many tapes).
(Score: 0) by Anonymous Coward on Wednesday May 11 2022, @02:52PM (7 children)
Also bear in mind that all drive platters are ganged to the same shaft. You don't get parallelism by mixing technologies with one different technology on a separate platter.
(Score: 2) by RS3 on Wednesday May 11 2022, @04:55PM (6 children)
Very true, but you could squeeze a 2nd head actuator (for the PMR-only surface) in the other corner and it might just work.
(Score: 2) by Reziac on Thursday May 12 2022, @02:57AM (5 children)
Maybe it's time to admit that the form factor is too small for today's data demands, and return to the full-height hard drive.
And there is no Alkibiades to come back and save us from ourselves.
(Score: 2) by RS3 on Thursday May 12 2022, @06:12AM (4 children)
I don't know if you're serious, but I'd be okay with it. For me, no matter what, I want improved reliability, including long-term # bytes lost. IE, if you lose 1 100TB drive, that's obviously far worse than losing 1 of 5 20TB drives.
(Score: 2) by Reziac on Thursday May 12 2022, @06:25AM (3 children)
I am serious. I too will take reliability over conveniently compact, and that "crap, lost the big disk" problem is precisely why I don't mind a stack of 3TB drives instead of one 26TB drive.
Of course, I'm not a datacenter that needs all 26TB online all the time, either.
And there is no Alkibiades to come back and save us from ourselves.
(Score: 2) by RS3 on Thursday May 12 2022, @06:21PM (2 children)
Datacenters RAID them together, and you can "cluster" RAIDS together, and "storage area network", etc., so individual drive size isn't important for total capacity. But then they divvy them up into virtualized machines anyway!
They care more about capacity per hectare. :)
"Hectare" was partially silly, partially serious, of course.
(Score: 2) by Reziac on Thursday May 12 2022, @07:22PM
Ha. Hectare may be a joke today, but you just wait!
And there is no Alkibiades to come back and save us from ourselves.
(Score: 0) by Anonymous Coward on Friday May 13 2022, @07:31AM
We also care about reliability and service uptime. The problem with overly-large drives is that the array takes forever to rebuild. Unless you are using enough redundancies or replicas, you will have a failure on rebuild and some data will be lost. Bigger disks can reduce your overall cost, but it depends on the specs and your fault probability projections, especially the URE. Running degraded for long periods of time or handling large amounts of data increase the probability of loss. That said, we are considering adding more of the latest generations of large drives. But we are only doing so while maintaining proper diversity to prevent complete failure.