Mass storage has rarely been a hotbed of innovative activity. Simply put, innovation in the hard drive field has consisted of making rust particles tinier and tinier and making the platters on which the rust is embedded spin faster.
The last few months have been quite different.
We’ve seen perpendicular drives being introduced which will eventually quintuple the amount of data that can stored in X amount of space.
More dramatically, silicon is beginning to move onto rust’s turf. We hear announcements of hard drives fortified with flash memory. We hear announcements of hard drives that will be just flash memory. For more extreme desires, we hear promises of an affordable, decent-sized RAM drive.
What are we to make of all this?
What is interesting about all this technology is that it can be and is being pointed in opposite directions: both in the usual direction of more and bigger, but also in the direction of less and smaller.
Perpendicular hard drives will eventually make 1 to 2 terabyte hard drives possible, but the technology is first being pointed at smaller devices: notebooks, media players.
Look at the silicon challengers, and you have to think small, simply because you can’t afford to do otherwise. A flash memory drive will be overall a bit better than good rust, but cost rather more, and could wear out faster (especially in a hybrid system). DRAM-based drives will be better than flash, but cost much more while being still bottle-necked by the speed of the hard drive interface. I might be willing to lay out $800 for an 8Gb RAMDrive, but I’m hardly going to pretend the average schmo will.
If Tarzan had to describe this, he’d say: silicon costs much, rust very cheap. Silicon very fast, rust very slow. SATA too slow for silicon, but too fast for rust.
It seems to me we need two developments to take place before we can square the circle:
1) We Need A Silicon Option: An interface meant for rust will never be fast enough to make silicon happy. Even 600Mbytes/sec (SATA III) is way too slow to handle DRAM well.
What can we do? Up to now, there really hasn’t been an alternative, but now there is: PCI Express. In theory, one could have a PCI-E x16 channel that could transfer data at 4 gigabytes/second, though I suspect people could live with somewhat less than that.
Current PCI Express motherboards don’t have nearly enough PCI-E channels to support another x8 or x16 device, but future generations will have many more channels to play with.
2) Recognize The New Divide In Storage When we look at new ways to store things, we need to realize that there is a new data divide for the consumer PC, and no one method can handle both.
What kind of divide am I talking about? There is computer data, and there is media data.
Computer data consists of applications and work done with them. In most cases, especially for home PCs, this only amounts to a few dozen gigabytes at most.
Then there is media data: music and movies. This is what chews up the rust, and only rust for the foreseeable future will be able to affordably handle it.
Granted, for those of you who already have media servers, this is a “DOH,” but this is going to dawn on everyone else the next few years.
A terabyte may sound like a lot, and it certainly is for typical home computer data, but in a few years, it won’t be a whole lot for media data. In a few years, a terabyte will only be able to store about 50 HD-movies.
Another important quality of media data is that it is inherent slow, and it doesn’t improve with speed. I don’t care how cyber you are, you aren’t going to start watching 90 minute movies in 3 minutes even if your hardware lets you. True, you might like to transfer that movie from one place to another very quickly, but not if that costs you $50,000.
Yes, there will eventually be a consumer market for 1 to 2 terabyte hard drives, but only to the extent Joe Sixpack wants to store all his entertainment on his hard drive. This is something PC makers will very much want him or her to do, and probably will become mainstream one way or another in the long run (if not necessarily in a PC). In the short to medium term (say within the next three years), thinking everybody is going to do this is quite iffy.
If you don’t want a media server, you don’t need a one terabyte drive, and odds are you won’t want to pay for one. On the other hand, if you can get the hard drive space you do need in a much smaller package, you’d probably like that, especially if it’s cheaper.
I think we’re going to see this divide emerge big-time the next few years: new drives for the computer data, rust for those (and eventually, only those) who need humongous storage.
A New Hybrid
We face an environment the next few years where we really need to speed up the computer data, but no single option is terribly desirable. If it’s cheap, it’s too slow. If it’s fast, it’s way too expensive.
We need to do two things to improve the situation.
First, we need to give ourselves a silicon option and make PCI-E a silicon option for delivering data.
Once we have enough broadband, then we need to be able to affordably fill it. Pure DRAM has speed to burn, but will cost way too much. Flash memory is more affordable, but it will have to get at least a magnitude faster and somewhat more durable while getting cheaper. This will be tough to do.
One option that hasn’t been considered yet is a marriage of DRAM and a hard drive, using a PCI-E interface. Take a little hard drive made possible by this perpendicular stuff and put it on a card. Provide one or two user-fillable DRAM slots, and make them (a little) programmable, in the sense of “I want A, B, and C” in cache all the time. If you want a near instant boot, you could set the cache up for that; if you have higher priorities, you’d set them. If you don’t have strong preferences, you’d let the cache work like a regular hard drive cache, except it would be a lot bigger.
The memory part of the device would be powered by the normal PCI-E current available while the machine is still plugged into the wall, with a battery big enough to power the device for a save whenever the power goes off. Users could configure other save options, too.
Such a hybrid wouldn’t offer all the advantages of a solid-state drive, but it wouldn’t break the bank, either. If you told people they could boot in a few seconds for an extra $80 or so, I’d think a ton of people would take that, even if it did nothing else, and those who want more can pay for it.
Another, maybe even better possibility is to simply keep one or more RAM slots powered when the machine is turned off, and make any other necessary changes to allow at least some main memory to serve as a bootable, persistent RAMDrive capable of feeding data at multi-gigabytes per second. No doubt it would make mobos more costly, but that would be even cheaper than a hybrid PCI device.
Realistically, even assuming everyone who needed to get involved did, we wouldn’t see something like this for a couple years.
But wouldn’t it be worth it?