SoylentNews is people
SoylentNews
Seagate: 100TB HDDs Due in 2030, Multi-Actuator Drives to Become Common
Seagate is on track to deliver ~50TB hard disk drives by 2026, ~100TB HDDs by 2030, and 120TB+ units early next decade, according to the company's recently revealed product and technology roadmaps. To hit capacity targets, Seagate will have to adopt new magnetic recording technologies. To ensure the high performance of its future drives, the company plans to leverage its multi-actuator technology more broadly. This tech doubles the performance of its hard drives, and it could become a standard feature on some of the company's product lines.
[...] Today's [heat-assisted magnetic recording (HAMR)] media is expected to enable drives featuring 80TB ~ 100TB capacity, according to developers. But, for 3.5-inch HDDs with a ~105TB capacity and 5 ~ 7Tb/in2 areal density, new ordered-granular magnetic films will be needed as grains will get very small and tracks will get very narrow. But ordered-granular media is expected to be a relatively short stop before 'fully' bit patterned media (BPM) technology comes into play with an 8Tb/inch2 areal density.
[...] A straightforward way to increase the [input/output operations per second (IOPS)]-per-TB performance of an HDD is to use more than one actuator with read/write heads, and this is exactly what Seagate is set to do. Using two actuators instead of one can almost double throughput as well as IOPS-per-TB performance, which is tremendously important for data centers. Furthermore, doubling the number of actuators also halves the time Seagate needs to test a drive before shipping, as it is faster to inspect eight or nine platters using two independent actuators, which lowers costs.
Previously: Western Digital to Use Microwave Assisted Magnetic Recording to Produce 40 TB HDDs by 2025
Seagate to Stay the Course With HAMR HDDs, Plans 20 TB by 2020, ~50 TB Before 2025
Seagate Plans 36 TB HAMR HDDs by 2022, 48 TB by 2024
Related: Toshiba Announces 16 TB and 18 TB Microwave-Assisted Magnetic Recording (MAMR) Hard Drives
(Score: 1) by vali.magni on Wednesday March 10 2021, @07:25AM (11 children)
As capacity increases I'm more concerned with errors (bit rot, etc.) and failure rates. Are there any projections around these parameters?
(Score: 5, Interesting) by ledow on Wednesday March 10 2021, @08:12AM (10 children)
Every Seagate drive I've ever purchased has failed miserably way under the "MTBF" average.
That's from enterprise down to desktop, used sparingly or for things like CCTV, SAN RAID, offline NAS, server RAID, etc. etc.
In my current workplace, that's dozens of drives all failing with the first few years of having them.
I've yet to rebuild a RAID array successfully that was based on Seagate drives (and, sure, it could be said that you shouldn't build an array with all identical drives, but when you buy them new and they come populated with them, you'd expect them to hold out at least a couple of years of non-intensive usage). Every time, while rebuilding, I get a cascade of failures and the array is just trashed.
As I see them, I just replace them with something else (anything else!).
I honestly thought the days of "Yeah, but Brand X drives always fail" was left in the days of DOS, the IBM Deskstars, Maxtor etc. but no. Everything I've ever purchased or populated with Seagate drives, whether server, home NAS, small business NAS, desktop or storage modules for blade servers or SAN unit, has failed miserably. Replace the drives with the appropriate WD equivalent and they just keep spinning (two drive failures in 7 years versus 40/50 in the first year alone of Seagate).
That said... I'm far more interested in SSD nowadays. HDD is dead, but they keep trying to flog it. You can't keep increasing capacity (and thus backup, rebuild etc. times) without some increase in speed, but HDDs are pretty much where they're ever going to be in terms of speed.
(Score: 0) by Anonymous Coward on Wednesday March 10 2021, @09:06AM
The MTBF isn't the median of the distribution generated by the function. If the MTBF has been calculated correctly, around 36% of the product population will survive to that time. It also doesn't hold in situations without a CFR. It is not surprising you haven't had a single product survive to their MTBF, just as it isn't a surprise that many claim they haven't seen a failure yet.
(Score: 1, Interesting) by Anonymous Coward on Wednesday March 10 2021, @01:54PM (6 children)
SLC with its 100,000 rewrites had an implicit lifelong guarantee (and explicit too, in most cases).
MLC with 10,000 and slower but 2x space made sense in many places.
TLC with 3,000 and doubly slower, with 1.5x space of MLC? Hmm... maybe, for those thrifty souls out there.
QLC with 1,000 and yet again slower, with 1.3x space of TLC? Seriously?
PLC with 300, turtle slow, with 1.2x space of QLC??? Dude, WTF???
I would not have complained about fools playing the stupid game and winning the stupid prizes, if not that good flash devices all but DISAPPEARED from the market. Today you are hard pressed to find yourself a MLC drive, and SLC has all the fine qualities of unobtainium.
Building myself a home datacenter and dancing around it with replacement drives day after day after day, is not the life I desire to be forced into. And still it seems there is dedicated effort to make storage devices as unreliable as possible; to get home users to give up and gift all their data to "the Cloud" and the Wise Masters who own it?
(Score: 3, Interesting) by takyon on Wednesday March 10 2021, @02:30PM (1 child)
8-bits-per-cell OLC, anybody?
Luckily, TLC is not disappearing from the market just yet. DRAM and "SLC cache" can make up for the speed loss. The write endurance is not needed for most consumer devices. The capacity increases from 3D stacking can also cover up these issues. 1 TB drives can have a bigger cache than 256 GB, and so on, and the user is even less likely to fill a larger drive hundreds of times.
Back when planar NAND was reaching its limits, I remember seeing research about rejuvenating NAND to increase endurance. It's probably this:
https://silvertonconsulting.com/blog/2012/12/04/heating-nand-brings-it-back-to-life/ [silvertonconsulting.com]
https://ieeexplore.ieee.org/document/7524803 [ieee.org]
3D NAND allowed Samsung, Micron, et al. to kick the can down the road on fixing NAND's issues for at least a decade. They are going to shoot for no less than 512 layers, using string stacking. QLC is probably fine for most people, as long as the drive size is large, but TLC could stick around because QLC is not much cheaper. For something stupid like PLC, unpowered data retention may become an issue.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 0) by Anonymous Coward on Wednesday March 10 2021, @03:58PM
TLC is proven to lose data in a few years if left unpowered. And its being powered does not do magic, it does periodic rewrites... eating into that "not needed" endurance limit.
SLC, on the other hand, does not deteriorate for a decade at least (had one drive sitting in a drawer for that long, all contents in perfect state afterwards).
(Score: 1, Interesting) by Anonymous Coward on Wednesday March 10 2021, @03:26PM (3 children)
TLC is the sweet spot, probably. 1000 writes is enough if you don't flog it horribly, and it will probably be obsolete before it wears out. I got a new NVMe TLC drive to replace my old SATA system disk a month ago and despite copying over my whole OS, I still don't even have one full write. At this rate it will last decades even if I only get the minimum 300 writes (but I don't use it for swap space). TLC is still readily available - easier to find than a non-SMR mechanical consumer drive, and with less of a price premium too. I have never seen a MLC drive in NVMe configuration, though I have a couple in SATA form. SLC is for data centers where write lifetime is essential. Too expensive otherwise.
One thing that I like about SSDs is that the minimum price is so low. You can get a cheap SATA SSD for $20, and SD cards for $5. Abuse them all you want, it's disposable, just throw it out when it dies. Mechanical storage never costs much less than $50 new, and at the low end isn't even all that much of a price advantage, because manufacturing that high precision machinery is just inherently pricey. Mechanical drives scale up, but SSDs scale down.
(Score: 0) by Anonymous Coward on Wednesday March 10 2021, @04:00PM (2 children)
Cue the dancing with replacement drives. DO NOT WANT.
(Score: 0) by Anonymous Coward on Thursday March 11 2021, @02:05AM (1 child)
All drives fail. Used to be if you got five years out of a mechanical drive you were doing well. Of course until the late 00's most cases were basically hard disk torture chambers without airflow or vibration damping. I have a few mechanical drives, the oldest of which is a 2008 1TB WD Green (these are known for longevity as long as you manage their idle spindown). It's starting to make bearing noise and probably only has a year or so left. This is the best case scenario for a mechanical drive - but attainable if you have a healthy drive environment.
My main OS installation is from 2007 (the benefits of rolling release) and it's been through four drive migrations. You have to do this whether you like it or not.
It seems like we're getting into a place where drive capacity is increasing faster than the need for it. There's a good chance that my new SSD lasts over a decade before it's either obsolete or fails. There's a nonzero chance that my next one outlasts me.
(Score: 0) by Anonymous Coward on Thursday March 11 2021, @02:03PM
Problem is not the system drive but archive ones. A proper drive should NOT fail while lying in a drawer.
(Score: 2) by leon_the_cat on Wednesday March 10 2021, @11:50PM
Totally agree these high capacity drives are junk as actuator/head failure is way too high. Possibly another reason they are going to double up on this.
(Score: 0) by Anonymous Coward on Tuesday March 23 2021, @11:00AM
I have a 500GB Seagate Barracuda that has resulted in corrupted ext? partitions multiple times, a 3TB that would corrupt everything past the first ~100GB with bad sector errors (can't tell if it's firmware or some other issue.) and then a 1TB and 1.5TB that are both still running bug-free 10+ years on. (Due for an upgrade, but thanks to Fry's last couple of years I haven't had a place to buy them since I don't order online or have a credit card.)