News Seagate's Mach.2 Technology Doubles HDD Performance, Microsoft Jumps Aboard

bit_user

Polypheme
Ambassador
Sooo really what they are doing is putting two smaller hard drives in one enclosure. That is all. They should of just said that.
Not quite. They rotate on the same spindle. And they obviously share a controller and interface. It really is only the actuator arms that are independent.

There have been prototypes (I'm not sure if any released products) that had two sets of actuators & heads for accessing the same platters. In that case, you would literally have double the bandwidth, or the ability to read & write at the same time. The "deal-breaker" likely has to do with calibrating the two sets of heads.
 

NeatOman

Distinguished
Jun 2, 2009
73
1
18,635
Would be nice if there is a RAID controller built into the drive when/if this goes to consumers. But maybe one step at a time. Still much cheaper than two 8TB $1500 SSDs vs tiered SSD and HDD if you don't need the SSD performance that much. Double the HDD performance per rack is much more than it's not lol.
 

alextheblue

Distinguished
Would be nice if there is a RAID controller built into the drive when/if this goes to consumers. But maybe one step at a time. Still much cheaper than two 8TB $1500 SSDs vs tiered SSD and HDD if you don't need the SSD performance that much. Double the HDD performance per rack is much more than it's not lol.
This probably won't see a regular consumer model. Cost is too high. Most consumer applications either don't need a ton of IOPs (mass storage on a secondary large HDD), or benefit from LOTS of IOPS but don't generally need as much capacity (OS + applications, which is better served by an SSD).

Professional users probably won't care either, since tower space is rarely an issue, and you can RAID a ton of cheaper conventional drives and/or bolster with SSD caches. The only prosumer application that might make sense is a NAS/media server situation, if your NAS is flexible enough (since you're effectively doubling the number of HDDs).
I clicked, hoping to see performance figures; should’ve known better.
Performance is rather similar to two 7200 RPM enterprise HDDs, but occupies the physical space of one. That should give you a pretty good idea, since that is already a known quantity (as in, awful).
 

bit_user

Polypheme
Ambassador
The only prosumer application that might make sense is a NAS/media server situation, if your NAS is flexible enough (since you're effectively doubling the number of HDDs).
Yeah, especially for a small or medium-sized office. Also for database servers and other scenarios where IOPS matter and you wouldn't necessarily just use SSDs.

Performance is rather similar to two 7200 RPM enterprise HDDs, but occupies the physical space of one.
I think it's more like they cut the stack of platters in half and put different actuators on each. So, the best-case scenario is that you might get twice the random IOPS out of it, but I'd expect the sequential transfer rate to be as bad as half of the equivalent conventional drive.

Going back and looking at the article, I see the claim:

“Based upon that analysis, we’ve found we are getting close to twice the throughput and IOPS, which are the improvements we expected to achieve with the MACH.2 technology,” says Ogus.

In an IOPS-heavy workload, where much of the time is spent seeking, I would expect throughput to nearly double. However, for more sequential workloads, the only way I see this increasing throughput is if it had more platters than a conventional drive. Either that, or conventional hard drives don't use more than one head at a time, which I doubt.

As for extracting real-world performance out of it, I think a lot has to do with how data is split between the two halves. I can think of cool ways to split it, to suit different specialized use cases.


BTW, hi!
 
Last edited:
  • Like
Reactions: alextheblue

JoBalz

Distinguished
Sep 1, 2014
101
42
18,640
https://www.tomshardware.com/news/seagate-hdd-harddrive,8279.html

I was just speaking to a friend about how hard drive articles from a decade ago sound the same as the ones coming out now. Funny thing that we're now seeing this article. The link above is from a Tom's article over 10 years ago, speaking on the same topic, vendor and technologies.

HDD has been around for a long time so the technology is fairly mature. That doesn't mean there's not room for improvement. In this case the technology had been attempted in the past but for various reasons it didn't prove feasible at the time. With other improved technologies and production techniques, these HDDs are now economically feasible to produce them since HDDs, As with any mature technology, we won't read about HDDs as often as we do in the rapidly improvements in SSDs.
 
  • Like
Reactions: bit_user

bit_user

Polypheme
Ambassador
https://www.tomshardware.com/news/seagate-hdd-harddrive,8279.html

I was just speaking to a friend about how hard drive articles from a decade ago sound the same as the ones coming out now. Funny thing that we're now seeing this article. The link above is from a Tom's article over 10 years ago, speaking on the same topic, vendor and technologies.
Except it's really not the same.

I alluded to this, above. In the article you cited, every hard disk sector is reachable by 2 heads. In this case, each sector is only reachable by a single head. That article does a pretty good job of explaining why the old approach could not be carried forward. I think the main issue is that mutually calibrating multiple heads per platter being just to difficult and fragile.

There is one notable thing about that article:
one commenter proposed a question: why don't hard drive manufacturers do something like add a second set of read/write heads to increase performance rather than turning up the spindle rate to dangerously high levels?

Good question! We went directly to Seagate for the answers.
Oh man, I miss the days when reporters would bother to visit the forums and actually talk to companies and try to get answers - not just parrot the latest press releases. Well, at least @PaulAlcorn still does, on occasion.

Based on what I assume has happened to the site's readership numbers + the burying of the forum posts on the article pages, it's not exactly surprising that we get no love (or much in the way of journalistic heroics).
 
  • Like
Reactions: alextheblue

bit_user

Polypheme
Ambassador
for various reasons it didn't prove feasible at the time.
Except I think this technique was technically feasible, but just didn't have such a strong value proposition as it does in the days of 10+ TB drives.

One thing to keep in mind is that HDD transfer rates are increasing at a square root of their capacity. This has a few interesting implications. For one, the bigger the drives get, the longer it takes to do things like rebuild a RAID. At some point, it takes so long that the risk of another drive failure becomes significant. Over time, this will put downward pressure on the number of disks in a RAID, and I think data centers are probably near or possibly past the point of even using conventional RAID.

Another problem, and the one highlighted in the article, is that while you can physically pack more data on a drive, getting it on & off starts to become the limiting factor in how much you can actually put on a single drive. So, I wonder if there's any chance we might see RPMs start creeping back up. Another possibility is data centers increasingly embracing 2.5" form factors for mechanical HDDs.
 

bit_user

Polypheme
Ambassador
Would be nice if there is a RAID controller built into the drive when/if this goes to consumers. But maybe one step at a time.
Please explain how this would work.

The way I see it, if you're talking about RAID 1, then the only thing it saves you from is one of the actuators failing. However, if the controller, spindle bearing, or spindle motor fails, then you're still SOL. So, I don't see much value in that.

If you're talking about RAID 0, then probably a single-actuator drive of the same capacity would deliver that performance.

Am I missing something, here?
 

seanwebster

Contributing Writer
Editor
Aug 30, 2018
191
68
10,690
Except it's really not the same.

I alluded to this, above. In the article you cited, every hard disk sector is reachable by 2 heads. In this case, each sector is only reachable by a single head. That article does a pretty good job of explaining why the old approach could not be carried forward. I think the main issue is that mutually calibrating multiple heads per platter being just to difficult and fragile.

There is one notable thing about that article:

Oh man, I miss the days when reporters would bother to visit the forums and actually talk to companies and try to get answers - not just parrot the latest press releases. Well, at least @PaulAlcorn still does, on occasion.

Based on what I assume has happened to the site's readership numbers + the burying of the forum posts on the article pages, it's not exactly surprising that we get no love (or much in the way of journalistic heroics).
Just ask! Feel free to reach out whenever. I'll be sure to get you the right answers to anything you need storage. :)
 
  • Like
Reactions: bit_user

gaborbarla

Distinguished
The whole mach.2 naming implies that the existing technology is mach 1, which implies faster than sound... Clearly something is wrong with this naming. I hope never to own another HD again.
 

bit_user

Polypheme
Ambassador
Just ask! Feel free to reach out whenever. I'll be sure to get you the right answers to anything you need storage. :)
Thanks for the offer. It's one of those cases where I'm quite happy to be proven wrong. :sweatsmile:

Sadly, I missed the Intel Optane AMA. I would've asked them about error correction. So far, all of my SSDs have end-to-end error correction. I wonder if they see value in offering that for non-datacenter Optane drives (i.e. such as those targeted towards enterprise, professional, and commercial applications), or is there some reason why it's less relevant for Optane products?

Maybe you could try to squeeze in a line of inquiry about that, next time you cover Optane related news, product launches, announcements, etc.

Also, on a related note, I wonder why us home NAS builders can't buy Ryzen Pro CPUs. AMD literally won't sell them to us and no stores or online retailers even carry OEM versions of these chips. It's really annoying, when I can even buy a motherboard that supports them, but the only way to source the chips is to get used, older models on ebay that people have pulled from enterprise desktops.

Anyway, thanks & happy holidays to you & yours!
🎅
 

bit_user

Polypheme
Ambassador
The whole mach.2 naming implies that the existing technology is mach 1, which implies faster than sound... Clearly something is wrong with this naming. I hope never to own another HD again.
Cute, but dude... it's just a name.

Gillette Mach 3 razors do not enable you to shave at 3x the speed of sound.

71jefO4jh6L._SY355_.jpg

What say you about that?
 

alextheblue

Distinguished
BINGO, I have never had good luck with Seagate HDs, always have found WD much more reliable and longer lasting in my experience.
That's been my experience with their consumer drives. I'd be willing to bet their server drives are an order of magnitude better. Then again, I've had issues with some WD models too here and there - especially first-gen Raptors. Great, now you've got me rambling. Might as well finish: IBM was pretty solid except for a particular series of Deskstar (nicknamed "Deathstar" IIRC). The Hitachi drives that came after were great too, and my Samsung drives were bulletproof. Maxtor consumer drives were the ones I had the WORST luck with on average. Even worse than Seagate, if you can believe it.
Also for database servers and other scenarios where IOPS matter and you wouldn't necessarily just use SSDs.
Indeed. That's why I specified prosumer (and the OP was talking about potential consumer drives). There's quite a few server applications, naturally.
Either that, or conventional hard drives don't use more than one head at a time, which I doubt.

BTW, hi!
I believe this is how most if not all current HDDs work, yeah. It's why performance doesn't scale with platters (drives like this Mach 2 or perhaps other rare examples aside). Probably nearly impossible for them to perfectly align data from platter to platter and/or ditto for the heads, and keep it all in alignment. Or something of that nature. Even if all that worked perfectly, the performance would be more susceptible to degradation due to fragmentation, even compared to a conventional spinner.

Hope you're doing well, Bit.
 
  • Like
Reactions: bit_user

bit_user

Polypheme
Ambassador
I believe this is how most if not all current HDDs work, yeah. It's why performance doesn't scale with platters (drives like this Mach 2 or perhaps other rare examples aside). Probably nearly impossible for them to perfectly align data from platter to platter and/or ditto for the heads, and keep it all in alignment. Or something of that nature.
This is a good point. I had previously wondered why performance didn't seem to scale with platter count, but you're probably right about too much difficulty keeping all the heads aligned. The arm could warp too much, over the operating temperature range - it's probably something like that.

Hope you're doing well, Bit.

🎅
 
  • Like
Reactions: alextheblue