Western Digital unveils 20TB OptiNAND hard drive, pledges 50TB to follow

Western Digital has announced a "breakthrough in storage that works differently," in the form of a new architecture combining traditional platters with solid-state flash: OptiNAND. Adding flash to traditional mechanical hard drives is not a new concept. Western Digital announced its first work on the concept back in 2011 after …

[TeeCee]

Sounds to me like, in this case. "boosting performance" actually means "getting performance back to where it was before we started playing silly buggers with areal density".

Holding the metadata and indices in "iNand" may well reduce the need to refresh adjacent tracks, but I can't see it speeding up R/W performance like, say, the traditional "flash cache" drives do.

Having said that, the increased density usually produces a hike over lower capacity platters and we'll probably get to see this, rather than losing it to packing overhead as is more usual these days.

[eldakka]

> but I can't see it speeding up R/W performance like, say, the traditional "flash cache" drives do.

I think it might not be a raw bitrate speedup, it sounds like an overhead reduction that effectively, from a users point of view, results in more data being read/written in the same time period. Sorta like when PCIe went from 8b/10b encoding to 128b/130b encoding with v3 (for PCie prev3, transferring 128b of 'user' data would actually require transferring 160b of raw data, PCIe3 changed this to requiring only 130b of raw data to get that 128b of user data). Since that extra metadata is in NAND instead of bits on the disk, it doesn't take up data read/write bandwidth from the platters to retrieve/store this metadata.

Actually, some of it is mentioned (peripherally) in the article:

At the same time OptiNAND is claimed to offer boosted performance by reducing the number of track interference refreshes required as well as the number of write-cache flushes

All of those operations take time that could be being used for reading/writing user data to the platters instead of pausing user read/write operations for these maintenance tasks.

[Martin an gof]

I think you are both correct. My reading is that increasing density also increases interference, which in the past has always meant an efficiency drop due to the need for more frequent refreshes, so this technique is a way of - possibly only partly - mitigating that drop with the next bump in density.

The thing that is confusing me though is that claim that previous generations needed refresh perhaps as often as every ten (or so) writes, but kept track in DRAM which presumably is lost on every power cycle. I realise that a drive used for online storage may rarely be switched off but when it is, how does that work? Does the drive need to do a massive maintenance cycle, writing every occupied sector, on every startup, "just in case"? Does a "normal" shutdown cause the data in DRAM to be written to a reserved area of the disc? Is a permanent on-disc copy of DRAM data kept to guard against unexpected power loss?

Any of the above would result in additional performance penalties which could be mitigated by using flash instead of DRAM.

Is the flash also used in the traditional "cacheing" role, which would provide additional speed-up for certain workloads, or is that not where this drive is aimed?

Whatever, my home NAS is stuck with 2TB drives and a full resilver of one of those can already take a day and impacts performance (not that performance is a huge problem at home). 20TB strikes me as the sort of capacity where the performance hit of resilvering, possibly taking a week or more, could get tedious.

M.

[Zbig]

Why so modest

Why so modest, Mr Siva?

"We were first to try to con our customers into buying SMR-based "NAS" drives not fit for purpose and almost got away with it".

"We have pioneered pushing 7200 RPM HDDs as "5400-class" drives to people who have made a conscious decision to go with slower RPM for energy, noise and/or vibration considerations".

"We have invented the marketing technology to switch out SSD components for inferior ones in the middle of the product's lifecycle without changing its name or letting anyone know".

Seriously, F off, WDC

[Anonymous Coward]

Re: Why so modest

sorry for the 'metoo' upvote, not sure about the ssd components switch (wasn't it samsung, in recent news?), but it felt wrong to spoil all those lovely 24 upvotes so there, Mr Silva of WD, would you care to reply to those spurious, public accusations?

...

Thank you sir, I couldn't expect less in all that (ample) time to provide a cohesive, bullshitting yeah-but-no-but-yeah-but-no-but response!

[Sorry that handle is already taken.]

Re: Why so modest

It was Samsung most recently, but it was WD only a few days before that.

And Adata and Crucial, just before WD.

[Vometia has insomnia. Again.]

WD claims WD is doing awesome things

Sounds legit. :| Wondering how this factors into their shingled stealth cheapertech innovationalism...

[biolo]

I can't decide if this them doing something new with the HDD storage side of this product, or if what they are doing is essentially shingling on the HDD and using the flash to overcome the worst performance hits of that. If it is different from shingling I wonder if the two techniques could be combined for a further performance boost.

[Anonymous Coward]

Don't know, Don't care

Ages ago, when I built a PC, I went with WD because they were the best on the market.

Now, I have no idea if any of their claims are true.

But they've fiddled so much with their HDs and SSDs behind their customers' backs that even if the claims are true today, there is no assurance that they will be true tomorrow. WD has a long way to go if they want to get off my 'don'f trust, don't buy' list.

[ITS Retired]

20 & 50TB

That's a lot of eggs in one basket, if/when something happens to the hard drive.

I was recently looking for a hard drive for use by just the operating system. The smallest I could find locally was 1TB. Yeah, yeah, I know I could partition it to something reasonable. But still, hard drives are mechanical devices. Separate, smaller hard drives would lessen the pain, if something happened. Not everyone is an expert in recovery. Back-ups? Too many people don't understand properly, the how and worse, the why. Bigger is not always better.

Don't get me started on TV/monitors in restricted spaces.

[Throatwarbler Mangrove]

Re: 20 & 50TB

I'm afraid I don't understand your point. What is the value in having smaller partitions or smaller drives? With modern filesystems, significant corruption basically doesn't happen, and having multiple drives presumably increases your chances that mechanical failure will happen, increasing the odds of some sort of data loss. If you want redundancy, buy two large drives instead of one and create a mirror set. Ultimately, there's no substitute for backups, though. While most people may not understand how to do backups in the traditional sense of separate physical media, I think many people at this point have at least some awareness of cloud backup solutions (OneDrive or Google Drive, to name just two), which help protect the data users care most about. While the traditionalists among us may turn their noses up at the idea of cloud backups, for most people, backing up to the cloud is still an improvement on what they would have otherwise.

[katrinab]

Re: 20 & 50TB

A 1TB SSD costs about £100. Smaller drives are available for a lower price. Why would you consider a mechanical drive for something that small? Especially a system drive?

[Anonymous Coward]

Re: 20 & 50TB

being petty I was going to point out that my 2 year old 480 samsung evo-something still costs around 70 quid, so there's no way 1Tb costs 100...

WRONG! 1TB ssds are, indeed, between 80 - 104 quid.

... strangely enough, samsung 860 EVO 1tb is 106 quid (on sale, sure) and 500Gb equivalent is... 99 quid.

[katrinab]

Re: 20 & 50TB

You have chia miners to thank for that.

You can get a 1TB QVO for about £80, but it doesn't count in my book.

[DS999]

I remember

When people were saying the same thing about 20 GB drives. Probably there were people saying it about 20 MB drives back in the 1970s.

[katrinab]

Re: I remember

One thing that hasn't changed since then, or at least since the 1990s, is spindle speeds. If anything, the average spindle speed has got slower because the 10,000 and 15,000 rpm drives have been mostly replaced with ssds.

This means that reading or writing one track of data takes the same time now as it did then, but there are a lot more tracks on each platter, so reading or writing an entire platter takes a lot longer.

Sure, there is more data on each track, so data transfer speed is faster, but it is only increasing by the square root of the increase in data.

There are also more platters in a drive, that increases capacity as well, and does increase speed by the same amount.

[Bartholomew]

Re: 20 & 50TB

> if/when something happens to the hard drive

It is filled with helium so a 5 year lifetime. If you have not backed up and moved your data to the next storage device before the helium leaks away you have lost it anyhow. Do not get me wrong, helium filled drives are fantastic for the data centre, but not for long term home storage for data.

And SSD's if powered off will not keep data as long as you might expect*, because it can not detect and repair failing data without power. The workaround for SSD storage, if being used to archive data, is to operate them as hot as possible while writing data, and when powered off keep them as cold as possible. But heat will shorten the overall operating lifetime of everything, ultimately due to the Arrhenius equation (which for tech roughly translates as "every 10°C increase in temperature will reduce the life of electronics by half").

(* ref: Search for "JEDEC SSD Specifications Explained Alvin Cox, Seagate Chairman, JC-64.8" and read pages 26 and 27 "SSD endurance classes and requirements").

[Bartholomew]

Re: 20 & 50TB

> * ref: Search for "JEDEC SSD Specifications Explained Alvin Cox, Seagate Chairman, JC-64.8" and read pages 26 and 27 "SSD endurance classes and requirements").

The raw data used in the JEDEC presentation was provided by Intel, and it is available, just difficult to find for some odd reason.

For SSD's and data retention, heat is good and cold is bad while writing the data and the reverse is true while not powered up, which if you think about trapping a few electrons between insulators from a physics point of view kind of makes sense.

[Martin an gof]

Re: 20 & 50TB

I was recently looking for a hard drive for use by just the operating system

Your average Joe isn't really looking for massive amounts of local storage these days - all their data is "in the cloud". The days when a laptop with a 4TB HDD was a necessary purchase so that you could download all those videos from your phone, which probably only had 4GB or 8GB of internal storage are long since gone.

This drive is obviously not aimed at the "home" user, unless it is that class of home user that will buy something like a MyBook in order to create a "cloud" at home.

Regarding desktop-type builds, I really can't remember the last time I fitted spinning rust as the boot disc. You are right, you definitely don't need 1TB for the OS, and if you buy a 1TB drive and partition it you have all the same performance problems, and loss of the drive means losing all the partitions, so all the restore problems too.

For perhaps 10 years now, maybe more, I've been fitting a low-capacity SSD for booting and adding higher-capacity HDDs for storage where necessary. Indeed I had some computers - using networked storage - where a 60GB SSD was the only local storage, giving plenty of room for the OS with some left over for local user data.

Byte-for-byte, small capacity SSDs are less value-for-money (perhaps £25 for 120GB NVMe against £110 for 1TB SATA) I suppose, but that's the only real argument against using them in this way.

M.

[Sub 20 Pilot]

Re: 20 & 50TB

Over a large number of years I used WD HDD's in an external case as part of my backup strategy. Apart from a couple of times I dropped them onto a solid floor they all worked well until they got full and were changed.

In the last 3 years or so I have had 3 fail on me for no reason ( luckily I never rely on one ) May be luck of the draw or more than likely becoming crapper to save money which seems endemic in the ICT world - PC's, phones, printers - all shite-ier quality as time moves on. ( Couldn't think of the correct multiplier word for shite..)

[ShadowSystems]

Old fart rant inside...

I recently purchased a WD 16Tb desktop external drive for doing backups & archival storeage. I was nonplussed when it turned out to only have 14Tb of formatted space & only 12Tb of user space available. I swear to Cthulhu's chocolate frosted knickers that it should be illegal to market a drive as $CapacityX in the large print, only to qualify that as $CapacityY in the small print & a further reduction to $CapacityZ for the actual user space. If I wanted a 12Tb drive I would have bought a 12Tb drive. I bought a 16Tb drive in the naieve belief that it wouldn't lose a quarter of the advertised space to something entirely beyond my control. If you have to format it to use it, and that formatting reduces the drive by an amount that is not inconsiderate, then you should be forced to market the drive as it's formatted size instead. I'll stop bitching & moaning now. It's time for my dried frog pills...

[Yet Another Anonymous coward]

Re: Old fart rant inside...

It's not upto WD how much space your OS filesystem reserves for it's own use.

If I install VMS and reserve 50% of the space for previous file versions that isn't really the drive makers problem

[ShadowSystems]

At YAAC, re: drive space.

Sorry I wasn't clear on that. This drive came from WD already formatted for Windows file storeage. So the manufacturer knew exactly how much space the drive really offered to the user. They knew it only offered 12Tb of user space, but still advertised it as a 16Tb drive. How much does a 4Tb 7200RPM HDD cost these days? Because I feel they owe me a free one to make up the difference.

*Hands you a pint & taps tankards*

Enjoy a pint on me to make up for the miscommunication on my part.

[Pascal Monett]

Re: Old fart rant inside...

Ah, but it had 16Tb, as in tera-bits.

And, as we all know, there's 8 bits to a byte.

Plus there's the fact that marketing can't be arsed to calculate the kilobyte at its true value of 1024 bytes, it uses 1000 because if marketers knew how to count they would be beancounters.

[Bodge99]

Re: Old fart rant inside...

Sorry, but a kilobyte **IS** 1,000 bytes.. You are thinking of kibibyte (1,024 bytes).

This spec. has existed since 1998.

B.

[vincent himpe]

Re: Old fart rant inside...

Can we cut this crap once and for all ?

That kibi stuff should never have been created !

It has been understood, all the way back from the early days of computing that kilo, in computing world, means two to the power of 10

case in point :

If i buy a dimm (ram) module that is 1 gigabyte it is two to power of thirty bytes = 1073741824 bytes. not 1.000.000.000 bytes.

If i buy a microcontroller that has 32k of ram and 64k of rom it has 32768 bytes of ram and 65535 bytes of rom. not 32000 and 64000

There is not a single semiconductor manufacturer that uses that kibi-wibi bullshittery. Even the flash chips in that SSD use two-to-the power-of metrics.

They should have left kilo and mega and giga alone instead of renaming it wibbly-pippi-wippi.

What they should have done is create something like FakeMega FakeGgiga, or MarketingGiga or even simplier : Megalies, Gigalies , Teralies.

How big's that drive ? " 1 Fake Gigabyte aka 1 Marketing Giga. or the 1 Gigalies.

Marketing.. never trust any marketing specs.

[vincent himpe]

Re: it's crap

it's not base8 math. it's base 2 maths. To control 1 kilobyte you need 10 address lines to generate 1024 unique binary codes. For Mega this becomes 20 address lines. Giga is 30 address lines.

Since each control line is binary it has two possible states.

hence :

kilo = 2^10

mega = 2^20

giga = 2^30

[Snake]

Re: it's

Yes, I suck at explaining things as I'm walking with a phone

[Norman Nescio]

Re: Old fart rant inside...

In contrast, in the telecommunications industry, one kilobit per second is 1,000 bits per second, not 1024 bits per second. A good old ISDN 64 kilobit connection transfers 64,000 bits per second, not 65,536. This has caught people out occasionally.

It is unfortunate that certain parts of the IT industry appropriated the SI prefix kilo and redefined it to mean 2¹⁰ instead of 10³. The standard use of kilo- to mean multiplication by one thousand pre-dates electronic computing by some margin. Nevertheless, for some people, kilo- naturally means multiply by 1,024, so it is up to the reader to understand the context and translate/interpret accordingly, which is not ideal, but is the way of the world.

[vincent himpe]

Re: Old fart rant inside...

Ahahaaaah. LOL. SI -facepalm-

Let me tell you a joke about SI

The international System of Units aka SI. from the International Bureau of Weights and Standards (BIPM)

Hold it right there.

International System <=> SI

International Bureau of Weights and Measures <=> BIPM

It is in reality:

Systeme Internationaux (sorry, don't know how to create the french accents)

Bureau International de Poids et Mesures

So we've all been using that wrong all these years. "The international system" "SI"

yeah, mon lecteur DC utilise du CD .. ( disque compacte) (courant directe) as they say over there. (my Compact Disc player uses Direct current)

Second joke. The SI specifies that MULTIPLIERS are UPPERCASE , and fractions are lowercase.

so:

Multipliers:

T = Tera

G = Giga

M = Mega

Fractions

d= deci (1/10)

p = pico

n = nano

Hold it. What about Deca (x10), Hecto (x100), Kilo (x1000) ? They are multipliers too. Ah, mais oui, those are lowercase. WHY ? Because we're french. We like exceptions. Our language is built on exceptions.

Ok, but what with the single letter prefix rule, doesn't that cause an issue with deca (x10) and deci (1/10) ?

yes, so for deca we will use TWO letters : da

So we have a "systeme international" where we use a single character to indicate a multiplier or divider. We use uppercase for multipliers and lowercase for dividers, but deca hecto and kilo will use lowercase (despite being multipliers) and deca gets two letters (despite the standard using one letter everywhere else).

You've been a wonderful audience , next up a monkey riding a monocycle while reading shakespeare.

[Anonymous Coward]

Re: Old fart rant inside...

unfortunately, this is the trick the (...) have used on hdds and legally got away with for the last... well, I'd say... since mid-1990s (before then I wasn't into THIS size...)

[pwl]

Re: Old fart rant inside...

upvoted for “ Cthulhu's chocolate frosted knickers “

i am going to have to find a way to subtly include that in a conversation sometime soon

[Bitsminer]

Re: Old fart rant inside...

Try reading the label on the box. Or the footnote on the specification sheet:

One MB is equal to one million bytes, one

GB is equal to one billion bytes and one TB

equals 1,000GB (one trillion bytes). Actual

user capacity may be less due to operating

environment.

[Sub 20 Pilot]

Re: Old fart rant inside...

But that doesn't make it right - to be able to lie about the usable capacity. Even though we know this has happened for decades it does not make it right and it should be stopped as being an outright lie.

If I sell you a house with three floors and it turns out that only one of them can be used, but in some microscopic print in a 1000 page EULA or similar, I say that the number of floors may not be usable by the purchaser then you would rightly have a case against me. No difference in my opinion.

[Anonymous Coward]

"breakthrough in storage that works differently,

unlike all previous, regular breakthroughs that worked the same as pre-breakthrough breakthroughs.

[Kevin McMurtrie]

It's not shingled

Due to the high density, flash storage needs to carefully track writes that may require adjacent track refreshes at intervals approaching 1 cycle.

[Anonymous Coward]

Re: It's not shingled

No it doesn't