80TB HDD with HAMR coming

[snortingfrogs]

Showa Denko K.K (SDK) is showing their new technique for much larger HDD's.

 

HAMR stands for Heat-Assisted Magnetic Recording.

 

HAMR represents a recording method in which magnetic film is locally heated at the time of recording. This technology has been developed to solve the “magnetic recording trilemma”: difficulty in simultaneously meeting the three requirements of fine-particle structure, resistance to thermal fluctuation, and ease of magnetization. Compared with the recording density of approx. 1.14 Tb/in

² for HD media based on conventional magnetic recording methods, it is said that HAMR-based HD media will achieve recording density of 5-6 Tb/in² in the future. Provided that the same number of disks are used, it is estimated that a 3.5-inch HDD will achieve storage capacity of approx. 70-80 TB per unit.

 

The 80TB drives are a way off though, but 20TB and larger using the technique will arrive from Seagate later this year.

 

Source: https://www.sdk.co.jp/english/news/2020/37882.html

 

This is really good for us horders and finally HDD's starts get bigger as we've been stuck on 8-12TB for way too long.

[SeriousDad69]

Unless they Quadruple(or more) the R/W speed IDGAF if they make a 500TB HDD.

It's getting to the point where it's like trying to drain/fill a swimming pool with a straw.

[RejZoR]

We haven't been stuck on 8-12TB. It's just that it's a very niche segment coz people generally don't need this much space. And their prices rise rather unreasonably. Not many people are willing to pay over 400€ for 16TB HDD (Seagate Exos drives are 16TB already). Though if you look how you only get like 2TB SSD for that price, 16TB HDD is huge and cheap.

[mr moose]

Man that's a lot of porn.

 

8 minutes ago, SeriousDad69 said:

Unless they Quadruple(or more) the R/W speed IDGAF if they make a 500TB HDD.

It's getting to the point where it's like trying to drain/fill a swimming pool with a straw.

 

Most users don't try to fill and read the entire hard drive, they just need a few gigs here and there.  So long as search times are stupid it might have a decent use as part of a  backup/archive system.

[SeriousDad69]

Just now, mr moose said:

Man that's a lot of porn.

 

 

Most users don't try to fill and read the entire hard drive, they just need a few gigs here and there.  So long as search times are stupid it might have a decent use as part of a  backup/archive system.

I get that, it's just if you ever have to actually backup from it it's going to take like 4+ days at max read speed to read all 80TB. Multiple 12TB drives are the sweet spot until they find a way to increase read/write speeds and reliability. Never put all your eggs in one hard drive lol

[mr moose]

9 minutes ago, SeriousDad69 said:

I get that, it's just if you ever have to actually backup from it it's going to take like 4+ days at max read speed to read all 80TB. Multiple 12TB drives are the sweet spot until they find a way to increase read/write speeds and reliability. Never put all your eggs in one hard drive lol

 

How would having multiple drives be faster?  If you need to get all 80 TB's then you have to sit through the same amount of data transfer.  Also just on the eggs and basket issue, two of these gives you a redundant copy and 80TB of backup except in two 3.5" drive space.  how many 3.5" drives do you need to get 80TB with a backup copy on 12TB drives?

[SeriousDad69]

1 hour ago, mr moose said:

 

How would having multiple drives be faster?  If you need to get all 80 TB's then you have to sit through the same amount of data transfer.  Also just on the eggs and basket issue, two of these gives you a redundant copy and 80TB of backup except in two 3.5" drive space.  how many 3.5" drives do you need to get 80TB with a backup copy on 12TB drives?

If you have say, ten 14TB HDDs in like a RaidZ2 all of the data will be spread out over all of the drives, allowing you to push around or even over 1GB a second in some situations. Compared to one 80TB drive pushing no more than 100-150MB a second at the best of times.

Mirroring the drive to another with like Raid1 might double your read speed in some situations, but if one drive fails you still have to spend 4+ days copying all of the data over to the replacement(and you're going to be shitting your pants the entire time, paranoid about the other drive failing while being pegged at 100% usage for so long).

The worst I would personally do is two 14TB CMR drives in Raid1 with my uber important data backed up to somewhere that has a client side encryption feature like Backblaze or MEGA or something so no one can snoop through your files while they're in the cloud.

[leadeater]

 

Still think the 12TB we've been buying recently is huge lol

[leadeater]

2 hours ago, SeriousDad69 said:

I get that, it's just if you ever have to actually backup from it it's going to take like 4+ days at max read speed to read all 80TB. Multiple 12TB drives are the sweet spot until they find a way to increase read/write speeds and reliability. Never put all your eggs in one hard drive lol

Realistically anyone buying these types/sizes of disks wouldn't be buying 1 or 2 anyway. It would take the maximum capacity of my 6 OSD node Ceph cluster from 2.3PB to 11.5PB, but I'd rather scale out than up for many reasons.

[SeriousDad69]

1 minute ago, leadeater said:

Realistically anyone buying these types/sizes of disks wouldn't be buying 1 or 2 anyway. It would take the maximum capacity of my 6 OSD node Ceph cluster from 2.3PB to 11.5PB, but I'd rather scale out than up for many reasons.

Yeah, 80TB drives are definitely just for like Amazon, Google, Microsoft etc. Although If they were stupid cheap like $1,000 each I could see myself being tempted enough to do something stupid like trying to build a 1PB NAS with them. Realistically, they're probably not even going to be available retail.

[Doobeedoo]

Great to see such capacity though, but being HDDs they'll have to improve IOPS/TB as well, more actuators and other tech. For cold storage, datacenters, or mass storage.

Though SSDs will definitely overtake in consumer space over time.

[Bombastinator]

Reaching and passing tape capacity plus (slow?) random access.  May finally kill tape maybe.

[BuckGup]

4 hours ago, Bombastinator said:

Reaching and passing tape capacity plus (slow?) random access.  May finally kill tape maybe.

Nah tape had 50TB cassettes 10 years ago. They are here to stay for awhile 

[mr moose]

8 hours ago, SeriousDad69 said:

If you have say, ten 14TB HDDs in like a RaidZ2 all of the data will be spread out over all of the drives, allowing you to push around or even over 1GB a second in some situations. Compared to one 80TB drive pushing no more than 100-150MB a second at the best of times.

So you need 14 drives to get the equivalent storage of 2 including a backup.  I am not familiar with ZFS what happens with a drive failure running z2, with 7  more drives you have 7 x times the chance of a drive failure.

 

8 hours ago, SeriousDad69 said:

Mirroring the drive to another with like Raid1 might double your read speed in some situations, but if one drive fails you still have to spend 4+ days copying all of the data over to the replacement(and you're going to be shitting your pants the entire time, paranoid about the other drive failing while being pegged at 100% usage for so long).

The worst I would personally do is two 14TB CMR drives in Raid1 with my uber important data backed up to somewhere that has a client side encryption feature like Backblaze or MEGA or something so no one can snoop through your files while they're in the cloud.

Either way at 80TB your still waiting a long time. 

[SeriousDad69]

5 hours ago, mr moose said:

snip

RaidZ2 can have two drives fail before data is lost. The odds of three properly housed drives failing within the same week are astronomically low. The way I see it is I would rather have eight to sixteen straws filling a swimming pool vs. one straw. And a one drive failure in a RaidZ2 with ten 14TB drives shouldn't take more than 12-16 hours to recover from.

Two 80TB drives in Raid1 = 50% storage loss, vs. ten 14TB drives in RaidZ2 = 20% storage loss.

[mr moose]

1 minute ago, SeriousDad69 said:

RaidZ2 can have two drives fail before data is lost. The odds of three properly housed drives failing within the same week are astronomically low. The way I see it is I would rather have eight to sixteen straws filling a swimming pool vs. one straw. And a one drive failure in a RaidZ2 with ten 14TB drives shouldn't take more than 12-16 hours to recover from.

Two 80TB drives in Raid1 = 50% storage loss, vs. ten 14TB drives in RaidZ2 = 20% storage loss.

I wouldn't raid them at all. 

[SeriousDad69]

1 minute ago, mr moose said:

I wouldn't raid them at all. 

Well yeah, if you just want to load it up with your steam library there's no problem there if it dies.

[mr moose]

2 minutes ago, SeriousDad69 said:

Well yeah, if you just want to load it up with your steam library there's no problem there if it dies.

I imagine these drives will start to replace tape for cold archiving.  (depending how long they can hold data in an off state)

[leadeater]

1 hour ago, mr moose said:

I imagine these drives will start to replace tape for cold archiving.  (depending how long they can hold data in an off state)

It's going to be a bit of a race, it's a two to three year cycle between each tape generation with the current (LTO-8) being released in 2017. I currently use LTO-7 which has a raw and compressed storage of 6TB/15TB at throughput speeds of 300/750 MB/s peak, average I get about 9TB-11TB per tape because not all data can be compressed etc.

 

LTO-8 has a raw and compressed storage of 12TB/30TB at throughput speeds of 360/900 MB/s peak. LTO-9 is projected to have a raw and compressed storage of 24TB/60TB at throughput speeds of 708/1770 MB/s peak and is due likely next year. HAMR HDDs are likely further away from that and also won't have nearly as high throughput speeds, but unlike tapes each disk is also a drive, 1:1 relationship unlike tapes which is Many:Few relationship.

 

So when 80TB HAMR comes around it might be competing with LTO-10 which is projected to do over 100TB compressed.

[RejZoR]

I have trust issues with the HAMR component. From what I read this tech needs to heat up the positions where it's writing to efficiently write to it. What I don't know is whether it needs to heat it up for a read too or not. Coz why would you only need it for write, but reading it can be done as is. And as if HDD's weren't unreliable enough due to their mechanical nature, they will now have additional heating element on its head that will be an additional failure point. And thirdly, speeds. SMR already affects write performance because it works in a similar fashion as SSD's. It can't just write directly, it needs to also pull data that's shingled on top of where you want to write. I wonder if HAMR uses SMR as basis and does heat assisted writing on top of it for even further density. Hitachi neatly avoided all that by using Healium and their drives are rare exception where they still have insane capacities, but don't use SMR which means their write speeds are not affected at all. So, if you want high capacity and high speed of writes today, you need to go with HelioSeal drives. It's gonna be interesting seeing how these drives behave in real world scenarios and how reliable they'll be coz of that heating element.

[mr moose]

9 minutes ago, RejZoR said:

I have trust issues with the HAMR component. From what I read this tech needs to heat up the positions where it's writing to efficiently write to it. What I don't know is whether it needs to heat it up for a read too or not. Coz why would you only need it for write, but reading it can be done as is. And as if HDD's weren't unreliable enough due to their mechanical nature, they will now have additional heating element on its head that will be an additional failure point. And thirdly, speeds. SMR already affects write performance because it works in a similar fashion as SSD's. It can't just write directly, it needs to also pull data that's shingled on top of where you want to write. I wonder if HAMR uses SMR as basis and does heat assisted writing on top of it for even further density. Hitachi neatly avoided all that by using Healium and their drives are rare exception where they still have insane capacities, but don't use SMR which means their write speeds are not affected at all. So, if you want high capacity and high speed of writes today, you need to go with HelioSeal drives. It's gonna be interesting seeing how these drives behave in real world scenarios and how reliable they'll be coz of that heating element.

Because the particles are so small that it is very hard to accurately align them with the read head and not have thermal issues mis align them.  Writing fine particles (achieving density) is the problem not reading.