SSDs are far lower cost-per-access, either reading or writing. And what good is your data when it is not being used?
In effect, all the valuable data ends up on SSDs, which will be the work-horses of storage. All the cold archives and junk data will stay on HDD to save costs.
As for enterprise SSD, ugh. Wrong idea. Look for nearline, and learn to use it efficiently. The two worst problems for SSD are:
- no-one gets taught what it is. They still use it like disk, which it is not even close to being. It is something new, distinct, and much better Everyone from the OS to the app writers needs to go back to school on this, and repeat the mantra "Flash is not Disk". Basic rules are write sequential, read random, and read up on "log structured file systems" and "journalling logs".
- the drives have FTL firmware which tries to pander to programmers who are mindlessly continuing to use the SSD as if it is disk. Causing needless complexity, wear, bugs, and distractions.
But if you do use Flash for what it really is, the rewards are great. Amazing performance, productive programmers, and rapid return on investment.
The postscript is that HDDs are no longer Disks like you think they are. But, that is another long story.
The 1000x is my recollection (I do recall seeing it, but as you noted, sometimes people are too optimistic -- on the semi side, we'll see if EUV ever hits production...), from a couple years ago, and might not be achievable given the current situation.
It looks like there's a pretty solid roadmap to 5-10x density using HAMR, then add in patterned media (I'm not sure how much more that could add) -- and maybe other approaches (such as magnetic holography or plasmonics, but those are too far out)
Agreed, you started your post about 15k. Then you meandered, as did I.
The core issue is your claim of 1000x gain in HDD capacity. I did a search for roadmap articles. Nothing that optimistic comes up. Also, the last ones published are from 2012 which is just before the flying height problems became critical. That is rather like looking at a CPU GHz roadmap just before Pentium hit the heat wall and GHz scaling ended. If you can't fly lower than about 5nm without excessive friction then getting order of magnitude density improvement is going to be very tough.
Do you have any pointers to specific, recent roadmaps on HDD density?
The only catch lies here is, if WD wants to take premium price for this new product then it will get lost in the battle between the solid-state and conventional hard drive. Even though 15000 RPM is a remarkable achievement they will not be able to sell it a premium price, Conventional HDDs have still its application but if they start charging higher rates people will start using Flash drives in place of that.
You could try reading attentively; I said there's not much future for super fast HDDs. A title reading "Last Gasp For 15K HDDs" would be accurate, unlike the misleading title.
My guess is that WD's enterprise storage customers probably still want more 15K HDDs, so they delivered; customers can be very conservative.
You can search yourself, and see that the HDD industry has a pretty good roadmap (e.g. HAMR until at least 2020, and good for at least mulitple terabits per in2, then add in patterned media). I wonder if flash will scale so well....
No roadmap I know of allows 1000x gain on HDD capacity. HAMR and shingling are maybe good for a doubling each. Helium helps a bit with flying, but the heads are already flying so low that wear rates limit the total transfers, you can't use the densest drives in high transaction rates (even if speed were interesting, which it isn't). You can increase stiffness but we are down to the last nanometer or two of available gap, after that the head will be surfing lubricant all the time and wearing really fast.
Those life extending technologies come at the cost of greatly changing the way the drive needs to be used. For example shingling works by writing multiple tracks in strict sequence, which is to say many megabytes. You either commit to using those disks in careful patterns (logging and archiving would be fine, as would be media files). There is some talk of using mapping techniques (rather like the space reclamation in the FTL of an SSD) to hide this but that is only likely to be interesting in low-stress uses like consumer PCs - if there are any consumer machines still using internal HDDs by the time shingling is on the market.
To be clear, these problems do not apply to the featured 15k drive. At its apparent capacity of 150GB/platter that would put its flying height in a wear free regime, and it is doubtless writing data in a classic magnetic approach with no track overlap and no heat assistance. The high performance drive market requires those conservative characteristics, which is why you don't see them anywhere near maximum density for the form factor. And, they probably are limited to little more than the capacity you see today because the techniques to get beyond would involve head wear or loss of carefree random writes, or both.
A 600GB SSD will be around $350, which is likely about the same as this 15k drive costs (look up the price of the 3.5" version). Not sure why you still think SSD is so much more expensive.
You can get slow, dense 7200 or 5400 HDDs significantly cheaper but they have limited value, cold data and warm archives. That 15k drive is something of a throwback, retrofitting for obsolete machines or stuck-in-the-past attitudes. All that spinning metal buys you about 5% of the transaction rate of a modern SSD at twice the power drain, 100x the weight, and similar price. And that is just for ordinary SATA-III drives.
It might be the last gasp for speed oriented HDD's (the 10K and 15K monsters such as these), but as long as storage demands climb, HDD's will have a major place, although they won't be a growth market.
Why? Because HDD's have a pretty solid roadmap to massive density increases (~1000x using HAMR and patterned media). Does flash have that?
A personal anecdote: I recently built an affordable, fast PC with a 256G Samsung 840 Pro SSD and a 1T WD Black. Yes, the SSD flies, but the WD Black HDD is no slouch, either -- the real difference in speed is less than raw specs would indicate (and it's nice to get 4x the storage for 1/2 the price).
And don't forget: you can never have too much storage or too many backups!
To be honest: SSD resp. flash arrays will take over the mass market even if they are slower than the magnetic media. But they are notoriously unreliable and will become even more so.
So: If you need a reliable storage media that will hold the data for the next 10 or 20 years, you currently have exactly 3 alter- natives:
1. magnetic tape (!) 2, MO drive (remember these ones?) 3. magnetic hard disc <EOL> (end-of-list)
So, the question is easy to answer: those who have to store big amounts of data for an extended period of time still need hard disc. Not in the consumer market numbers, but they will even be willing to pay the price.
A Book For All Reasons Bernard Cole1 Comment Robert Oshana's recent book "Software Engineering for Embedded Systems (Newnes/Elsevier)," written and edited with Mark Kraeling, is a 'book for all reasons.' At almost 1,200 pages, it ...