Seagate Achieves Milestone in HAMR HDDs: 16 TB Units Internally Tested

[Raevenlord]

Seagate has been hyping their new HDD density-improvement technology, HAMR, for some time now. The basis of HAMR (Heat-Assisted Magnetic Recording) in Seagate's ypcoiming HDDs is to increase platter density without having to resort to other solutions so as to increase HDD capacity (increased number of platters, increased footprint, etc). The company says that internal tests of 16 TB HAMR-based HDDs are going well, with expected market release to partners by 2019.

HAMR does keep up with compatibility for enterprise customers, being a drop-in upgrade for other HDD-based storage solutions - they're just higher capacity, higher-performance solutions that don't need any special treatment from deployers. The plan is to release 20 TB solutions by 2020, and a staggering 48 TB in the standard 3.5" form-factor by 2024. Seagate further states that HAMR-based drives far exceed industry-required reliability parameters, so the company is bullish on the attention its technology will garner once available to customers in general. The HAMR tech will be deployed firstly on the company's





"Seagate has begun running early Exos HAMR units through the full set of standard benchmark tests used to prepare and optimize each new hard drive product for deployment," says Jason Feist, Seagate's senior director of enterprise product line management. "Our testing has demonstrated the drives' compatibility for enterprise systems that are being used today. No system level changes are needed to run the HAMR drives in these evaluations, or to deploy them in customer environments."

View at TechPowerUp Main Site

 

[kastriot]

Begining of journey to the end of horizon.

 

[E-curbi]

Begining of journey to the end of horizon.

Seagate bought out LSI, who had previously purchased the early SandForce controller technology. What ever happened to those groundbreaking NVMe SSDs from Seagate everyone was expecting?

 

[natr0n]

More layers of failure points.

 

[DonKnotts]

The HAMR tech will be deployed firstly on the company's...

On the company's what?

 

[dj-electric]

Meanwhile, at 2018, creating an SSD of 1/10 the physical volume of that HDD with 16TB is just as easy as ever. But humanity doesn't want to advance properly towards it.

 

[Frick]

Meanwhile, at 2018, creating an SSD of 1/10 the physical volume of that HDD with 16TB is just as easy as ever. But humanity doesn't want to advance properly towards it.

We are though.

 

[dj-electric]

We are though.

People are being sold 2 flash chips, a controller and some 512MB ram to for 150$+ today. We're still some years from being fair on that thing.
I know i make that sound simplistic, but we're so far in the consumer side from actual affordable technology capability.

 

[FordGT90Concept]

So it only uses the laser to write? Also, are these drives helium filled or does the laser make that moot? 48 TB drives can't come soon enough...as long as the price is reasonable.

 

[Wavetrex]

So it only uses the laser to write?

No, it writes magnetically exactly as current ones, but instead of writing to a cold metal layer it writes to a hot one, so the bit flips much easier, requiring less energy.

With less energy needed also comes smaller writing coil, so a smaller bit, resulting in more capacity.
Brilliant !

Except, it was really, REALLY tough to get that laser miniaturized enough and precise enough not to damage the material while doing all this, perhaps hundreds of thousands of times in the same spot.

 

[Ruru]

Still not enough space for all the cat videos.

 

[Octavean]

So it only uses the laser to write? Also, are these drives helium filled or does the laser make that moot? 48 TB drives can't come soon enough...as long as the price is reasonable.

As long as the price is reasonably,.....!?!

Not a change, at least not in the short term with respect to the release date. Its going to cost "all the money".

Also you can't buy just one 48TB HDD. If it fails it could mean massive data loss. People would likely want to have at least two for RAID 1. Ideal at least four or five for RAID 5 or RAID 6.

 

[hat]

RAID 5 isn't smart for these mythical 48TB drives, let alone 16TB. It stops being smart around 2TB because of URE failures. In other words, a drive fails, you put in a fresh drive, and run into a URE when rebuilding the array... data gone.

 

[Readlight]

From my experience of Seagate, it loses big files and WD doesn't work longer than 10 years.

 

[Wavetrex]

I used Raid 5 for a long time but since growing beyond 3TB drives I gave it up.

Now a much more smarter solution for long-term storage, using SnapRAID to create "Parity Snapshots" on large amounts of data.
What this does is that it doesn't require the entire array to work perfectly to keep your data intact. Each drive is individual, and can even be taken out of the group easily without affecting the functionality of the others (but it still needs to be put back for sync and recovery, as the parity obviously depends on it)
Even more, it's possible to have from 1 to 6!!! levels of parity for monstrous number of drives (40 and up), all protected, up to any size.
UREs not relevant as well, as you would only lose the particular files affected by that dead block, and not the entire disk (or entire array!)

The disadvantage is that it's not real-time, you have to remember to sync regularly (or have a script running every day or so to sync).
It's also not adding the speed benefits of "hardware" RAID5/6, as each drive is still accessed individually - But on the other hand it's also much lower load on the drives, as during syncs only the parity drives get written, while the data drives are untouched. This is very different from normal RAID5/6 where everything is spread on all drives and there's massive Read/Write I/O torturing the drives on every read and write (and especially during recovery !)

~~~
With that said, UREs are not that dangerous today as they used to be before, because all modern drives have sector-level ECC, meaning things have to go REALLY bad in order to completely lose a sector.
Even more, marking bad/recovery is done cleanly by the controller, without blocking the disk on endless attempts to recover a bad sector. It just gets marked as bad, returns an I/O error but then moves on.

Overall, with all the tech that exists today, both hardware, firmware and software, it's very hard to actually lose data when you know what you're doing.
Sorry for the long text. Hope this helps somebody.

 

[AEKaBeer]

The issue I have with HAMR is I've been reading about it in physical magazines since I was 17. It's been 20 years of it being right around the corner but never delivered. I realise that Seagate's hand has been forced with WDs researchers on track with their own microwave equivalent to actually deliver HAMR drives, but until now it feels like HAMR is used to placate shareholders and fill slots for tech journalists.

 

[Jism]

So they microwave the data onto the platters these days?

 

[xkm1948]

Given the industry are shifting to QLC SSD, I am actually more inclined to trust good old HDD. QLC is shit for both performance and reliability. I would take a good HDD over a QLC SSD any day.

 

[Static~Charge]

RAID 5 isn't smart for these mythical 48TB drives, let alone 16TB. It stops being smart around 2TB because of URE failures. In other words, a drive fails, you put in a fresh drive, and run into a URE when rebuilding the array... data gone.

I had that happen on an old 14-drive RAID-5 setup at work. One hard drive failed, I replaced it, and a second hard drive failed during the rebuild process. Ugh.... In hindsight, we should've gone with RAID-6 for some extra redundancy.

 

[AEKaBeer]

Seagate's approach is to use small lasers to heat up the platter at the point of writing, substantially increasing the available density. WD is looking to use microwaves to acheive the same thing. I don't believe they need to do much with the reading heads beyond software, they are super sensitive to magnetic fields as it is.

 

[Absolution]

Are HAMR/MAMR drives competition to SSDs in terms of throughput?

Did Samsung realize that they were on the horizon and decided to dump SSDs as fast as possible in the market?

 

[gmn17]

We just can’t get enough...

 

[FordGT90Concept]

Are HAMR/MAMR drives competition to SSDs in terms of throughput?

SSDs will always have a latency/access time advantage because of how platters work. Bandwidth, 12 TB drives are less than 250 MB/s so the best you can expect these drives to do is 1 GB/s. They'll be deployed via SATA though (because practicality) which has a 6 Gb/s peak as part of the SATA III standard. In other words, actual throughput will be less than the drive is theoretically capable of unless the launch of 48 TB drives coincides with SATA IV specification which removes that chokepoint.

 

[comtek]

They should sell them in pairs for mirroring. Can't afford to lose data that Big.