Why no other hybrid drives?

[Plester]

The Momentus XT seems to be an unqualified success, reviews are good, it seems to be selling like hotcakes, so why hasn't WD, Hitachi or Samsung responded?

 

[corkyg]

Could be they were not ready to. Agree about the Momentus XT - I have three of them.

But, I see it basically as a management business decision.

 

[Idontcare]

Why no other 10krpm consumer drives besides Raptor?

Could be that for the same reason Seagate decided not to pursue WD into the 10krpm space that WD has decided to not pursue Seagate into the hybrid drive space.

 

[taltamir]

The Momentus XT seems to be an unqualified success, reviews are good, it seems to be selling like hotcakes, so why hasn't WD, Hitachi or Samsung responded?

because it is too big a risk to invest in the technology which is expected to be obsolete REALLY soon...
how soon? well, 25nm SSDs are expected in about a month, and the next shrink is not far away... (less than half current cost).
Because it requires people with expertise and understanding of both spindle AND ssd drives.
Because any investment into it means manpower and money not being invested into developing pure SSD tech, or platter density (the remaining bastion of platter drives is that they are much much bigger, and cheaper).
Also, if they start competing in that field, everyone's prices (and margins) will drop.

 

[bunnyfubbles]

yeah, basically because most of the major players in the SSD world aren't any of the major players in the HDD world.

Sure, Samsung has had some functional options as early on as anyone, and Western Digital just recently jumped into the fray, but neither company offers a drive that can hope to compete with the likes of Intel or any of the major Sandforce players.

It really sucks too, because as Anand pointed out in his Momentus XT review, any disc based drive could benefit from just the tiniest bit of NAND by quite a bit.

Of course maybe that's all for the best because enticing hybrid options might very well have stunted early SSD growth and hurt development.

And when we really think about it, these hybrid drives really are only useful as the boot drives, so those operating with pure SSD + disc for storage really wouldn't benefit from simply hybrids like the Momentus XT anyways.

With SSD cost bound to go down and capacity/performance bound to go up, it just seems like too little too late.

 

[FishAk]

What is there to develop?

For files above 32k, an SSD is only twice as fast, or about the same as a HDD- depending on whither it's reading or writing.

Smaller files- especially 4k and less- are dealt with multiple times faster on an SSD than a HDD.

For the average amount of writes in a given consumer volume, what is the percentage that falls below 32k? Maybe I'm off my rocker, and the percentage is too high, but I think it's feasable to put that much nand on a hybrid drive.

This is a brute force Russian solution, but the drive logic could simply put anything under 32k on the flash, and the rest on the disk. It would give the economy of the disk where it does the best, and the speed of the nand where it really counts. There is nothing for the disk to "learn".

A drive like this would be simple for consumers, and it wouldn't become obsolete until HDDs become obsolete. Everything is already in place. There is nothing to develop- it just needs implemented.

 

[RaistlinZ]

Like Bunny said, regular SSD tech is advancing so fast there's little point to companies investing heavily into hybrid drives. Capacity on regular SSD's are climing and price/GB are falling at a decent pace.

Maybe if you're running a small netbook that can ONLY fit one drive, then a Hybrid might make senes if you're on a budget.

 

[DrMrLordX]

I would rather see platter-based drives with massive DRAM caches.

 

[FishAk]

I would rather see platter-based drives with massive DRAM caches.

Now you're talking about a battery to keep the memory alive, or filling it from the disk every time you power the system. Not to mention the cost.

I'm serious.

Looking at Windows files- which are probably the heaviest for the percentage of small files- it looks to me like about 1 or 2 percent fall under 32k. Of course, with multimedia files mixed in, the percentage of smaller files falls considerably.

3 percent of a 1Tb drive is only 30Gb. A 1Tb HDD with 48Gb of flash for some spare aria could easily sell for around $200. With controller logic that puts the smaller files in the flash, your talking about an economical and powerful drive. I think there would be a real market for a drive like this.

The controller that separates the storage location by file size is the key. You can't do this with your current drive setup. Perhaps all we need is some smart programmer who can implement this.

 

[bunnyfubbles]

Now you're talking about a battery to keep the memory alive, or filling it from the disk every time you power the system. Not to mention the cost.

I'm pretty sure he's not talking about permanent storage on the DRAM, he's talking about a massive buffer.

Just like how the NAND gets reset in the Momentus XT should you ever reformat or even defrag the drive, the DRAM would likely work the same way but be used more as an extra buffer to help alleviate random reads/writes moreso than sustained throughput.

I'm serious.

Looking at Windows files- which are probably the heaviest for the percentage of small files- it looks to me like about 1 or 2 percent fall under 32k. Of course, with multimedia files mixed in, the percentage of smaller files falls considerably.

3 percent of a 1Tb drive is only 30Gb. A 1Tb HDD with 48Gb of flash for some spare aria could easily sell for around $200. With controller logic that puts the smaller files in the flash, your talking about an economical and powerful drive. I think there would be a real market for a drive like this.

The controller that separates the storage location by file size is the key. You can't do this with your current drive setup. Perhaps all we need is some smart programmer who can implement this.

The problem here is that you're assuming people need 1TB worth of storage space for their programs and also have a need for a drive that will intelligently juggle the files around to maximize performance. However I seriously doubt the average needed capacity for performance critical files eclipses much more than ~120GB. Gamers would certainly like more due to the average size of games being relatively large, but even then we're probably looking at no more than 250-500GB of space for a large majority.

Fact of the matter is that we're not that far off from having reasonably priced SSDs that are that size, so there simply won't be a need to mess around with hybrid drives except for in niche scenarios.

There was a time and place for your hybrid 1TB drive with ~30GB of flash. But that was a year or two ago, not now.

As it has been said before, its getting to the point where the situation of trying to speed up disc drives is pretty moot with NAND getting faster/cheaper/higher capacity. So it actually makes sense that recent developments in HDD tech is to actually slow them down as "green" drives for the sake of reducing heat/noise/power and increasing reliability because most of these larger drives really make the most sense as a permanent storage solution, particularly of large media files, and used in tandem with balls-to-the-wall fast pure SSDs.

 

[FishAk]

There was a time and place for your hybrid 1TB drive with ~30GB of flash. But that was a year or two ago, not now.

I'm assuming we are still a few to several years away from $0.20/Gb for SSDs- which is still more than twice the price of current HHDs. I hope you are right, and I am wrong.

 

[taltamir]

There is nothing to develop- it just needs implemented.

how do you implement it? just shove in some nand into a platter drive and cast a magic spell to make it work?
you have to invest to develop it, the tech to do that will not magically materialize.

 

[FishAk]

how do you implement it? just shove in some nand into a platter drive and cast a magic spell to make it work?
you have to invest to develop it, the tech to do that will not magically materialize.

What's wrong, God? You don't believe in magic?

Ok, Ok, sorry, I couldn't help it.

After thinking more about this, maybe it doesn't even have to be about drives per se. Maybe there's a way to do this with pure programming- provided you have one of each drive type.

But I'm not magic, nor could I program myself out of a paper bag.

Maybe I'll just go over here and smoke a little more of this crack...

 

[alaricljs]

I remember reading of a ZFS feature (possibly not evne implemented) that would do exactly that, take some amount of SSD space added to the pool and improve the speed of often accessed files. Could even specify storage pools of higher importance.

Aha, yup - http://www.nerdblog.com/2010/03/zfs-nas-followup-ssd-is-amazing.html

 

[fuzzymath10]

With careful implementation, a hybrid drive could be a useful solution for folks who either don't want to deal with the issue of multiple drives (not everyone is comfortable with an 80GB SSD + 1TB hdd configuration), or who can't because of physical limitations (i.e. one drive location). One that can show one drive but cleverly manage the flash and rotational media in the background would be good since a lot of ppl might mismanage a two-drive setup whereas one hybrid solution could automatically compensate for "overflowing" from SSD to HDD.

My sister's laptop has been running an X25-M but because her music is starting to pile up she's running out of room and not everybody wants to tote a 2.5" USB hard drive everywhere. So I'm going to put a momentus xt in, and take back the SSD, and hope the performance loss isn't too horrible.

 

[taltamir]

What's wrong, God? You don't believe in magic?

Ok, Ok, sorry, I couldn't help it.

Heh, that was funny 😛
as long as its friendly jibes its ok, its not like we are insulting each other or anything 😛.

After thinking more about this, maybe it doesn't even have to be about drives per se. Maybe there's a way to do this with pure programming- provided you have one of each drive type.

You are right, it is possible with careful programming, but for the idea is to make a single drive for someone who doesn't or can't have two drives, and also to minimize cost (aka, the smallest amount of NAND). It will take a mighty lot of effort and expertise to program something like that AFAIK.

The real kicker is the time to return on investment, and the long term return... if you are an investor and someone says "lets invest in hybrid drive"... you start asking some questions. You quickly find out that this is supposed to be a transient tech that will be rendered obsolete in a few years. it is unknown if it will earn enough money to pay for itself first, if your research and development is delayed, it can be rendered obsolete before you even finish research and thus you only lose money...

on the other hand improvements to SSD tech will likely remain profitable until their patent expires (ensuring max return on investment), can be used for years to come, and if development takes extra time, thats ok, you can use it when its done.

Platter tech is already seeming more risky, but still has potential to last years, so again, desirable to research. Remember when LCDs first came out? all research into CRTs was halted immediately, even though CRTs were still superior for a few years at first, nobody invested in CRT development anymore.

If I was asked to put my money into a tech, it wouldn't be into hybrid HDD, too risky.

 

[corkyg]

To reiterate, . . . I see it as a management decision. Where to invest resources that have the best chance for improving the bottom line.

 

[ElenaP]

What is the expected lifetime of the Momentus XT flash buffer?

 

[kmmatney]

What is the expected lifetime of the Momentus XT flash buffer?

If I remember correctly, the buffer uses SLC flash, so it has a very high lifetime - like 100K read/write cycles (and of course that is well managed to be spread across the flash). That's one reason why the buffer is small - it uses expensive flash.

 

[taltamir]

If I remember correctly, the buffer uses SLC flash, so it has a very high lifetime - like 100K read/write cycles (and of course that is well managed to be spread across the flash). That's one reason why the buffer is small - it uses expensive flash.

this is pretty odd, because the really fast drives get their speed from parallelism, and 100,000 writes isn't that much if you have a really small buffer.
100,000 writes @ 1GB = 100TB, while 10,000 writes @ 120GB = 1,200 TB.
so, the more flash you have, the more TB you can write to it before it runs out of writes. A small buffer will incur higher usage.

 

[FishAk]

Ya, you guys are probably right, and really, I hope you are. SSDs are only going to get faster, cheaper, and bigger. We all agree on that. The only question is time.


I suppose it's just a marketing decision, and those making the decision believe they have a better chance making money with the new hot thing. Even with the technology already in place, they must not believe they can sell enough units before the spindle drive is history. I don't personally agree with them, but they have information and resources I don't, so they may well be right. Any resources needed to retool a production line comes at the expense of another project, and whichever project the Board believes will make more money in the end, will, and should win.


Seagate started their hybrid project several years ago, and it probably has or will pay off for them. Even if it doesn't make money in and of itself, it certainly will pay dividends in their conversion to solid state- which must happen if they hope to survive. The value of the experience they get with their XT- both technical and marketing- can't be measured by units sold alone.

as long as its friendly jibes its ok

Among the first threads I read before joining this forum was your debate with Flamenko concerning the proper way to set up an SSD in Windows. The debate ended with your opponent's forfeiture. Your posts show a very well rounded and insightful mind, not limited to computers. I respect that very much. Occasionally you come down on the harsh side while making your points, but I have never seen an instance where you actually insulted someone. I respect that too.

 

[DrMrLordX]

I'm pretty sure he's not talking about permanent storage on the DRAM, he's talking about a massive buffer.

A massive buffer would certainly be acceptable, though permanent or semi-permanent DRAM storage in a desktop would not be a major challenge.

In any case, relatively slow DRAM should be inexpensive to implement as a drive cache, just not as cheap as NAND. DRAM would be simpler to implement I would think.

edit: hmm . . .

http://arstechnica.com/hardware/news/2007/09/coming-soon-hard-drives-with-1gb-ddr-ram-cache.ars

Seems like an overly-complicated way to implement a DRAM drive cache.

 

[Cerb]

The controller that separates the storage location by file size is the key. You can't do this with your current drive setup. Perhaps all we need is some smart programmer who can implement this.

How big are directory indices (read quite often)? How big are your settings files? How big are writes to session files, and logs? How many reads and writes for searches are large?

If you look only at files explicitly opened by the user, you won't find it. Think of all the secondary reads and writes going on all the time, though.

FI, as a real example, consider an Autodesk Land Desktop project directory. After a few months, one can take minutes just to get a directory listing, on a rotational drive, due to the drive seeking like mad. Throw that FS info on a solid state chip, and backups might not take hours, nor finding an old project taking minutes.

The problem? You can put the entire project directory on a SSD, affordably, today (if you can afford a software license, a few hundred bucks for a drive is nothing). Soon, even students will find it affordable. In addition, since most people don't use most of the space they have, larger sizes, and wear-leveling schemes, will help make up for the shrinking write cycles on cheap and small flash. For those of us who use the space, rotational drives will be added.

What you propose would work, and I think you even underestimate how often small reads can be bottlenecks, but the next step is making rotational drives house almost entirely larger files, with the main drive in most systems being a SSD. In high-end systems, it's already a no-brainer, and within a year or two, it will be that way for midrange boxes (when >=100GB SSDs get <=$100), and anything mobile. HDD players are scrambling to catch up on pure SSDs, and R&D to use hybrids will take engineers away from that endeavor.

 

[taltamir]

Seagate started their hybrid project several years ago, and it probably has or will pay off for them. Even if it doesn't make money in and of itself, it certainly will pay dividends in their conversion to solid state- which must happen if they hope to survive. The value of the experience they get with their XT- both technical and marketing- can't be measured by units sold alone.

I agree, it probably will. I wonder if seagate is the only one who had the courage to take that risk a few years ago. Or maybe foresight to realize its not as big a risk as others indicate and that SSDs will take longer to "kill" the HDD than everyone is estimating. Maybe another company took the plunge a few years ago as well and we will soon see competing tech for the momentous XT. Although, starting from scratch today seems a too late.

 

[FishAk]

HDD players are scrambling to catch up on pure SSDs, and R&D to use hybrids will take engineers away from that endeavor.

You're right, and I think I do appreciate the penalty a disk must pay for the head seeking small files.


The question is time.

If SSDs become cheep enough soon enough, there will be no need for a HDD, except for specialty purposes- which, by the way will widen the price difference between the two drive types in the opposite direction.

If, however we are still 3 to 4 years out from that point, I think the big HDD players, who are behind the eight-ball on SSDs, could get a real boost in both profit, and solid state experience, by implementing a hybrid drive. If this is true, than implementing a large capacity drive that puts all the frequent seeks on some nand, but the bulk on high density platters, would go a long way in helping them get caught up with both more resources from the profit, and experience. Looking at an ATTO chart, one can see that for files above 32K, reads are “only” twice as fast on an SSD, but writes are about the same. Of course when you start mixing these reads up, and they become more random, than the SSD has a bigger advantage. This is where the intelligence of the controller can make a big difference.

SandForce doesn't produce anything but a license, from what I can see. What is stopping the big boys from making a deal with them, for the use of their controller? If they played this right, WD, Seagate, Samsung and the others have the advantage, in that they already produce the high capacity disks, and have a good name in the storage world. In a few years, platters will be extinct, but right now, the SSDs really can't compete with those platters above 32k or so, in the consumer world- considering price/performance. Instead, those disk makers are trying to play catch-up in the new solid state storage world, but are using inferior controllers that can't hope to compete with the real ones. To me, this seams like a recipe to alienate their customer base, due to the fact that their SSDs will be seen as low quality by the average Joe. If there is still time, this could all be turned around with a screaming fast, high capacity, low price drive. Yes, it would take resources from their main goal- getting caught up in the SSD world- but I think it has good potential to pay it all back on several fronts.


Perhaps it's already too late, and SSDs will cross that threshold sooner than I think. It depends on how long it would take to actually get the disks to market. But they already have a head start, in that they own the platter world- such as it is-, and there are proven SSD controllers on the market. They would have to buy the nand just like everybody else, so it's really just a matter of deciding to use the resources, and tool up an assembly line. I think the actual implementation would be relatively short considering that all the parts are already developed.