Multi-Actuation: Seagate working to double IOPS on Spinning Disks

[leadeater]

4 minutes ago, XenosTech said:

I'm honestly surprised they didn't make 10k drives cheaper for consumers by now too

Hands up who had WD VelociRaptors 

 

Biggest reason is because the difference is so minimal and SSDs came in like a wrecking ball and slapped both 10K and 15K disks in the face and said "bitch please, sit down".

[XenosTech]

Just now, leadeater said:

Hands up who had WD VelociRaptors 

 

Biggest reason is because the difference is so minimal and SSDs came in like a wrecking ball and slapped both 10K and 15K disks in the face and said "bitch please, sit down".

I swear those things were mini ovens

[Dabombinable]

20 minutes ago, leadeater said:

Hands up who had WD VelociRaptors 

 

Biggest reason is because the difference is so minimal and SSDs came in like a wrecking ball and slapped both 10K and 15K disks in the face and said "bitch please, sit down".

The access times though.....even early 10k-15K drives if they were good to begin with still have access times that wreck any current 7200RPM HDD. Regardless of the noise.

16 minutes ago, XenosTech said:

I swear those things were mini ovens

Its the same for Maxtor DX740-L they get so damn hot for 7200RPM its not funny.

[leadeater]

1 minute ago, Dabombinable said:

The access times though.....even early 10k-15K drives if they were good to begin with still have access times that wreck any current 7200RPM HDD. Regardless of the noise.

It was pretty clear from the get go once SSDs hit the market HDDs were going to be relegated to low performance mass storage. I think the only thing that stemmed the tide of change so quickly was the NAND shortages due to the flooding of major fabs, that really slowed things down and increased prices.

[XenosTech]

5 minutes ago, Dabombinable said:

The access times though.....even early 10k-15K drives if they were good to begin with still have access times that wreck any current 7200RPM HDD. Regardless of the noise.

Its the same for Maxtor DX740-L they get so damn hot for 7200RPM its not funny.

Forget the noise... You need to put those right next to a fan that's running at 100% so it doesn't die from the sheer heat

[Bit_Guardian]

9 hours ago, leadeater said:

Hands up who had WD VelociRaptors 

 

Biggest reason is because the difference is so minimal and SSDs came in like a wrecking ball and slapped both 10K and 15K disks in the face and said "bitch please, sit down".

I had Velociraptors!

[Doobeedoo]

Interesting, specially regarding IOPS increase. Reminds me something like raid 0 in one HDD kinda but different that it can do two completely different streams of read/write independently. Like you copy two different things and it won't take twice the time cause of seeking. 

But I wonder how it will store data, like mentioned it will be great when bombarding HDD to do multiple things at once. Mainly two things, best case scenario right. Cause you would wait 1 by 1 thing to copy, render etc.

But say I want to copy paste a large game for example on disk it self usually it needs to first read part then copy, can't do both at same time. So I doubt that for such activity it would act as two separate disks internally. Cause game would need to be stored on platters that are controlled by one head and copied to platters controlled by other head. Somethng like two physical drives fo max read and write to achieve. Or game will be scattered across all platters. And maybe if it was only activity given to the HDD at the time of installation. What if you gave it various things to do, would installation be set only for one set of heads then and other operations to other.

Guess we'll need to see. 

 

[Blackie Sheen]

18 hours ago, Syntaxvgm said:

seems like itll be loud 

 

Its for data centers so in doesnt matter

[Coaxialgamer]

2*0=0

[ARikozuM]

@seagate_surfer, any comments? Is any of this under NDA or can you provide more info?

[seagate_surfer]

@ARikozuM thanks for the tag. Happy to see discussions going on regarding one of the new announcements. Giving specifics about stuff like this while it's still being worked on gets tricky, I'm sure you can appreciate that the company doesn't want us going very in depth with stuff that isn't fully out on the market yet so we don't step on any toes. What we can say is the blog post linked by OP in the original post has all of the specs for it that we can discuss at this point. Tom's Hardware did an article on the release here:

A tidbit from that article: "Seagate's new design uses two sets of actuator arms that operate independently. Each carries eight heads. That means the drive can read or write from two heads at once, provided they are attached to different actuator arms. The drive can respond to two commands in parallel and one head can read while another writes, or both can read or write simultaneously."

One thing still in the works (so take it with a grain of salt) is the idea of combining this with new HAMR developments (Backblaze Blog here) to hopefully provide 4x capacity and double the performance of current market hard drives. 

[ARikozuM]

1 minute ago, seagate_surfer said:

@ARikozuM thanks for the tag.

Thank you for the quick response! I fully understand being unable to talk of things until the time is right. Unsure if you can answer these, but will Seagate be implementing a controller that can control how sectors are written to maximize the likelihood of needing both heads at once and will Seagate be able to combine this with SMR (shingled magnetic recording) as well? 

1 minute ago, seagate_surfer said:

One thing still in the works (so take it with a grain of salt) is the idea of combining this with new HAMR developments (Backblaze Blog here) to hopefully provide 4x capacity and double the performance of current market hard drives. 

Note to users: This is not the Backblaze HDD failure rate. The link goes to HAMR and what it is and how it helps the industry. 

[seagate_surfer]

15 minutes ago, ARikozuM said:

Thank you for the quick response! I fully understand being unable to talk of things until the time is right. Unsure if you can answer these, but will Seagate be implementing a controller that can control how sectors are written to maximize the likelihood of needing both heads at once and will Seagate be able to combine this with SMR (shingled magnetic recording) as well? 

Note to users: This is not the Backblaze HDD failure rate. The link goes to HAMR and what it is and how it helps the industry. 

Not sure on the specifics of the plans with the controllers or SMR yet. It hasn't even been designated yet whether the multi-actuator tech will be implemented as SATA, SAS, NVMe, etc., so things like this are still a bit up in the air. Have to imagine there will be some consideration taken to how different firmware/controller uses could effect different drive types engineered for different purposes. Would be cool to see a purpose-built drive concept like SkyHawk  (surveillance use like DVR, NVR, CCTV) for instance, which is all about prioritizing write cycles for massive blocks of high-quality video data without frame drops, and hit it with this kind of tech to beef it up. It's fun to talk about and see where it might go.

Thanks for making the point about the Backblaze blog bit, probably should've thought to specify that ourselves. Backblaze good. Backblaze friend.

Edited December 20, 2017 by seagate_surfer
addressed previous comment question regarding SMR

[Jito463]

17 hours ago, leadeater said:

Hands up who had WD VelociRaptors 

[WMGroomAK]

6 minutes ago, seagate_surfer said:

Would be cool to see a purpose-built drive concept like SkyHawk  (surveillance use like DVR, NVR, CCTV) for instance, which is all about prioritizing write cycles for massive blocks of high-quality video data without frame drops, and hit it with this kind of tech to beef it up. It's fun to talk about and see where it might go.

This is great information!  If anyone in at Seagate is open for suggestions that you could pas on, I would really love to see this as something eventually implemented in your NAS drives for something along the lines of a home media server setup...  I could imagine it would allow for some really great multi-user/access systems.

[TetraSky]

Honestly surprised this wasn't a thing until now... 

[Syntaxvgm]

2 hours ago, Blackie Sheen said:

Its for data centers so in doesnt matter

I don't know if you've noticed, but datacenter tech often becomes consumer tech

[leadeater]

2 hours ago, seagate_surfer said:

@ARikozuM thanks for the tag. Happy to see discussions going on regarding one of the new announcements. Giving specifics about stuff like this while it's still being worked on gets tricky, I'm sure you can appreciate that the company doesn't want us going very in depth with stuff that isn't fully out on the market yet so we don't step on any toes. What we can say is the blog post linked by OP in the original post has all of the specs for it that we can discuss at this point. Tom's Hardware did an article on the release here:

A tidbit from that article: "Seagate's new design uses two sets of actuator arms that operate independently. Each carries eight heads. That means the drive can read or write from two heads at once, provided they are attached to different actuator arms. The drive can respond to two commands in parallel and one head can read while another writes, or both can read or write simultaneously."

One thing still in the works (so take it with a grain of salt) is the idea of combining this with new HAMR developments (Backblaze Blog here) to hopefully provide 4x capacity and double the performance of current market hard drives. 

This should actually be rather useful for disk backup storage. Currently we are looking at replacing our backup storage and one of the issues with this type of storage is making sure it can deliver the required performance, that means ensuring a minimum number of disks which also means having to buy smaller disks to keep the cost down but that increases the $/gb.

 

The current system has 300 ish 3TB disks with each tray using double parity and 2 hot spares so the usable capacity comes out around 600TB and can deliver a sustained performance of over 1GB/s or 15,000 IOPS, we actually would like more IOPS than that so double actuator would be awesome. We're still looking at replacement options but it's likely to be very similar to the current considering little has changed HDD technology wise and our performance and capacity requirements haven't changed (thanks dedup ).

[Blackie Sheen]

On 20/12/2017 at 2:44 PM, Syntaxvgm said:

I don't know if you've noticed, but datacenter tech often becomes consumer tech

Well what I meant is that it doesnt really matter right now if its expensive since they currently aim at data centers. It will probably trickle down  to consumer hardware when they improve and simplify the manufacturing techniques in order to reduce costs but probably not as the tech is right now

[bcredeur97]

why dont they put the other set of heads on like the opposite side of the platter... having them right next to each other seems like a nightmare waiting to happen to me

 

edit:  or you could have two read heads on every disk if you could put them on an opposite corner of the drive... that would actually make the drive more reliable if doable... (doubles speeds when working... provides a failover should one stop)

[Bit_Guardian]

On 12/21/2017 at 4:48 AM, seagate_surfer said:

Not sure on the specifics of the plans with the controllers or SMR yet. It hasn't even been designated yet whether the multi-actuator tech will be implemented as SATA, SAS, NVMe, etc., so things like this are still a bit up in the air. Have to imagine there will be some consideration taken to how different firmware/controller uses could effect different drive types engineered for different purposes. Would be cool to see a purpose-built drive concept like SkyHawk  (surveillance use like DVR, NVR, CCTV) for instance, which is all about prioritizing write cycles for massive blocks of high-quality video data without frame drops, and hit it with this kind of tech to beef it up. It's fun to talk about and see where it might go.

Thanks for making the point about the Backblaze blog bit, probably should've thought to specify that ourselves. Backblaze good. Backblaze friend.

I would guess SATA or SAS at most to start with. This isn't going to double sequential throughput, so you're still well within SATA's max bandwidth spec.

[Ertman]

Imagine if they could do something crazy like have 3 actuating arms per plane per platter.  I don't see how it would work without using smaller platter and the engineering for it to determine which head is doing what task etc would be nuts. And this doesn't begin to address issues with interference.

[leadeater]

5 hours ago, Ertman said:

Imagine if they could do something crazy like have 3 actuating arms per plane per platter.  I don't see how it would work without using smaller platter and the engineering for it to determine which head is doing what task etc would be nuts. And this doesn't begin to address issues with interference.

Maybe design it so it's physically impossible for collisions and use 2.5" platters in 3.5" casing for the extra room. Needlessly costly though for the gain and SSDs would still be faster and eventually be cheaper.

[Dabombinable]

1 hour ago, leadeater said:

Maybe design it so it's physically impossible for collisions and use 2.5" platters in 3.5" casing for the extra room. Needlessly costly though for the gain and SSDs would still be faster and eventually be cheaper.

However you can at least recover data far more easily of a dead HDD's platters than an SSD's NAND chips.

[leadeater]

1 hour ago, Dabombinable said:

However you can at least recover data far more easily of a dead HDD's platters than an SSD's NAND chips.

True, if you're prepared to pay for it. Most SSDs now days firmware lock in to read-only mode when they get too degraded so baring outright failure which would more likely happen in the first 6 months of operation data loss on SSDs is fairly safe, safer than with an HDD.

 

Still I would like to see dual independent actuator arms and read/write heads away, HDDs actually scale out very well with performance and capacity and with good controllers and caching most of the time all flash isn't actually necessary. Double the IOPs of HDDs and you'll see HDDs easily stick around for the next 15 years if the projected capacity improvements come to be.

 

This really only applies to server workloads though, I think there is nothing that is going to stop SSD's total takeover of the home/desktop market because you'll never be at the required scale or have the proper storage controllers and write-back caching with compression with copy-on-write to negate all the latency pitfalls of HDDs.

 

A lot of people get lost in the awesome numbers of SSDs and arrays and forget that most storage workloads actually doesn't need those insane performance figures, networking will always be the limiting factor and it doesn't take that many HDDs to saturate network links so unless you need the IOPs you don't need SSDs.