WD Plots A Course To 40TB HDDs With MAMR

Status
Not open for further replies.

jossrik

Distinguished
Sep 26, 2002
1,547
0
20,460
330


I'm still waiting for inexpensive 5Tb drives or such, and by that I mean we've had 75$ 3Tb drives for years and years, higher density, the price goes up and up. Ya, they do sell 12Tb drives, but you pay so much more for them that you might as well get a NAS and fill it with cheaper drives, or even cheaper, build a NAS. When will these awesome drives trickle down for mainstream use?
 

bit_user

Splendid
Ambassador
It amazes me to ponder the level of precision in modern HDDs. I wonder if there are any other mass-produced mechanical devices with such precision.

What's more, in 5 years of continuous operation (some warranties are that long), a 7.2 kRPM drive will spin almost 19 billion revolutions, and still have sufficiently low jitter that you can seek to within nanometers of any location on disk, in about 1/100th of a second or less. These are true marvels of modern science, engineering, and manufacturing, IMO. And yet, they're frequently derided as spinning rust.
 

pjmelect

Distinguished
Jul 14, 2006
3,175
0
21,160
159
There is no way that I would buy a drive that uses Shingled Magnetic Recording, I imagine that these drives will be fitted in cheap laptops built down to a price for unsuspecting buyers. The performance of these drives will be appalling.
 

bit_user

Splendid
Ambassador

They're currently targeted mostly at backup and cold storage. They're actually fine, as long as your writes are sequentially.

If you try to use them as a general-purpose HDD, then you're correct that their appalling random write performance is too big of a liability.

For those who don't know, these drives have (typically 15 to 40 MB) zones that overlap and therefore must be written only in their entirety. If you want to write only a subset of a zone, then you must read-modify-write the whole thing. This is the main reason for their poor random write performance. And as the areal density increases, zones sizes will only go up (it varies from one drive to another).
 
The company isn't telling us what capacity the leading models will ship in, but given the capabilities of the new technology, WD could quickly distance itself from competitors.
You should be able to estimate that base on the "Capacity Growth Outlook" graph they provided, which shows PMR density leveling off over the next few years, and maximum MAMR capacities continuing to grow at roughly 2TB per year in their place. That would imply that there will be no major surge in capacity, but that MAMR will simply take over as gains with PMR slow. Their graph doesn't show MAMR hitting the 20GB mark until at least 2020, so if the first MAMR drives appear prior to that, we can assume their maximum capacity will likely be lower, maybe 16-18TB, and probably at a price of over $500, like their current 12TB drives.

I would also take their suggestion that this will enable them to produce 40TB drives by 2025 with a grain of salt. That same graph suspiciously needs to curve sharply upward during the 2020s to get there. It's likely just a suggestion meant to instill confidence in investors, and they can be reasonably sure that no one will hold them to it 8 years down the line. I also have doubts that they will continue to maintain the same lead in terms of cost over flash storage over that time span, especially when you consider that for high-capacity bulk data storage, you don't need the same level of performance or write cycles found in most existing SSDs. With the slow rate of advancement in hard drive densities, it seems quite unlikely that they will hold a 10x cost advantage over flash storage in 2025, and SSDs designed specifically for bulk storage might potentially even make them obsolete by that point.


One thing they could potentially do is make them into hybrid drives with a suitably large flash buffer built in. So, for example, the drive could have 128GB of flash memory, which could be written to quickly, and then the drive could write that data to the platters in the background, whenever it's convenient. Files with frequent random accesses could remain exclusively in the flash memory. So you could have your high capacity multi-terabyte drive, but you would also have much faster performance than a typical drive under most common usage scenarios.
 

AnimeMania

Honorable
Dec 8, 2014
294
7
10,815
13


They could take some of that tech and reinvent the inkjet and 3D printer heads.
 

JonDol

Reputable
Nov 30, 2015
141
4
4,685
0
@AnimeMania: I'd be happy with less than that, like improving the life time... I used my previous printer mainly for scanning things about less than once a week and it died after 7 years. I think that it is too short considering my usage, compared to other mechanical HDDs I have that still work after 14 years and which were used for long months in 24h/7 running machines.

Other than that, there are long years since the storage area (whether mechanical or flash) is the one carrying the most innovations year over year (of all the parts of a computer) so I'm a bit less amazed, I think, than our colleague bit_user, about these products reliability. A completely different approach vs the planned obsolescence we usually see with the printers... Add to that the impossibility of using non manufacturer (read low cost) ink cartridges and we'll have a very happy picture.
 

dudmont

Reputable
Feb 23, 2015
1,404
0
5,660
198
To quote the immortal Homer J Simpson, "Could you repeat that part of the stuff where you said all about the things?"
While this is vaguely interesting, I'm sure you lost 99.5% of the traffic on this site when you went "in-depth" on how the new tech works. Keep up the good work!
 

bit_user

Splendid
Ambassador

Longevity is rarely much longer than people are willing to pay for. Sometimes there are outliers, but if the market doesn't value a longer service life for the kind of printer you had, then they can't justify using more expensive parts, lubricants, and manufacturing techniques to deliver it and still stay in business.

BTW, we have a small HP printer at work from about 10 years ago. It's probably printed millions of pages, by now, and just keeps kicking.


Don't confuse innovation with execution. I didn't say they were the most innovative area of tech. Developments are usually very incremental and we probably don't even hear about most of them, but I'm marveled at just how far the accumulation of those tweaks has gotten us.
 

bit_user

Splendid
Ambassador

Adding so much flash is not a trivial cost item, and becomes non-viable if it pushes GB/$ below that of PMR drives.

How much you need and how well this would work are highly dependent on the amount of randomly-written data. For something like a frequently-written database, this drive would still perform like garbage. For a drive with very light-duty writes, the flash buffer would have little benefit over the large RAM buffers drives currently use to perform partial zone writes (not to mention OS-level write buffering).
 

bit_user

Splendid
Ambassador

True, but the standard for enterprise seems to be >= 10^15 bits per uncorrectable read. Since anything much worse would probably mean a non-viable product, I think they won't ship until it's at least that good.

Actually, with increasing density (and throughput), I'd expect this number to be rising. But more sophisticated RAID implementations could probably deal with even worse.
 


I was referring more to perhaps a few years in the future when such drives could potentially be common and when flash memory should cost less. By that point, that much memory might not have to add more than $20 to the cost of a drive. Lower capacities, such as 64GB could also be possible, of course. And the added cost wouldn't matter much, since this wouldn't just be making up for the deficiencies of that recording method, but also enhancing responsiveness of the drive. The flash would not just be acting as a buffer, but also as a place where frequently accessed or modified files could be permanently stored, with the drive managing this in the background. Already there's hybrid drives like the Firecuda that do something similar, albeit with only 8GB for now, so this would just be a logical extension of that. Such a drive could perform similar whether it used SMR or PMR, and could perform significantly better than either type of drive without a large flash buffer, while being much more affordable than a multi-terabyte SSD.

And I imagine they were referring to home use, and likely won't be writing to massive databases that might not fit in such a buffer.
 

popatim

Titan
Moderator
Large HDD cache is nice to have until you have an app that has to clear the cache frequently. Even Samsungs 8gb cache SSHD's are a nightmare to use in those cirmcumstances.

Anywho, Intel already addresses this with Optaine, basically its HDD Cache done right, if it catches on...
 

bit_user

Splendid
Ambassador

Vaguely interesting? I think & hope you're wrong about how little interest there is in the details. I really appreciated Paul's efforts in making this new tech more accessible. More importantly, it's in-depth content like this that will hopefully attract and retain readers who are interested in such depth.

The only minor quibble I'd have is with the slides from CMU. For motivated readers, they're fine. But many readers wouldn't know what to make of many of the plots. So, ideally (and I'm not saying this is realistic), they should be explained or else the relevant bits should be extracted and made into separate plots. That said, I don't think they actually hurt the article, as is.


I hope you're not being sarcastic. If not, then I fully agree. Paul clearly puts time & effort into these articles, which I greatly appreciate.
 

bit_user

Splendid
Ambassador

Even as flash prices drop, you're still fighting a tough battle to deliver a SMR hybrid drive at a comparable GB/$ as PMR, given that SMR only improves density by ~25%.

IMO, the only place for hybrid drives is in laptops that need lots of storage. There, you get additional benefits by keeping the HDD spun down potentially most of the time.

In a desktop, I'd use SSD for the boot drive and a plain HDD for bulk data.
 

dudmont

Reputable
Feb 23, 2015
1,404
0
5,660
198
Bit User and all others,
I read most every article and news story that is posted, with the exception of much of the vr related and those that deal with games that don't interest me. I pretty much never struggle to figure out what's being written about(sometimes it does take time for info to sink in, like the fact that PCIE 3.0x4 m.2 drives are not really faster, in day to day ops than fast sata SSDs, as an example), but this story made my eyes glaze over about half way through, while Paul was discussing how magnetism is used on various types of platter drives. And to be blunt, I felt a touch of levity was needed, after such a rough cut of meat was attempted to be digested(the news story). To go further, I indeed appreciate the efforts of all the Toms staff. Paul does much good reporting, even if his football team now has finally lost a game, so I hope he does not feel demeaned, just merely slighted that I didn't take more time to come up with a snappy remark about the technical complexity of modern and future platter storage.
Dudmont
 
Nice to see all that space available but, there is one major issue that has not been resolved yet: Reliability! With all that we need 100% reliable hard drives. Get a lemon, loose all your data. Also, like others said, HD price only goes up while the 500 gb / 1 TB don't even lower the prices. It is really expensive to have a hard drive copy of a backup. I am not telling that WD had more failing HD. Just in general, we need reliability form any company.
 

I can certainly see room for hybrid drives on the desktop. While it might be fine for a power user to manage what goes on their SSD and what goes on their hard drive, most users likely wouldn't want to mess with determining what things they store will benefit from the higher performance of an SSD, and what is fine to store on their hard drive. The majority of people would likely prefer the simplicity of having all their files stored on a single partition, and leave it up to the computer to determine which files are being accessed in ways that will benefit from being stored on flash memory.

And in many cases, this could be more efficient. As an example, some people have game libraries that fill multiple terabytes, and storing all that on SSDs could be needlessly expensive and probably rather wasteful. So, they end up with the bulk of their game library stored on a hard drive, and only a handful of their most played games on the SSD. But some games load data in ways that they don't even benefit much from being on an SSD, while others will benefit quite a bit. Short of performing thorough tests and shuffling games back and forth between drives, it can be hard to determine which is which. And even within a single game, some files may be accessed frequently in ways that benefit a lot from quick access times, while others will only be read infrequently in a largely sequential manner. A hybrid drive could look at these access patterns and automatically manage which files, or even parts of files, will be best suited to remain mirrored in flash memory.

Plus, a large buffer area could be used to accelerate the speed of files being saved to the drive. You could, for example, copy some video files to the drive, and they would be saved at SSD-like speeds to the flash memory, then continue to be copied over to the platters in the background when no other accesses are happening. And if the drive is powered down or otherwise loses power, it can still resume the transfer without data loss when power is restored. Additionally, those copied files could be left mirrored in the flash memory, to remain available at higher performance until the buffer area is needed for something else, at which point the drive can intelligently determine whether to remove it from the cache, or remove some other files that haven't been accessed for a while.

And again, any performance deficit that a drive technology like SMR might experience could be almost entirely avoided in the vast majority of usage scenarios, since the writes could be stored to the flash memory before getting written sequentially to the drive platters at a later time.
 
Status
Not open for further replies.

ASK THE COMMUNITY

TRENDING THREADS