Slashdot Mirror
Next Wave Of Hard Drive Tech: Perpendicular Recording
angrytuna writes "New serial technologies are set to replace standard SCSI and ATA (Advanced Technology Attachment) interfaces over the next two years, even as hard-disk drive manufacturers prepare for an entirely new form of bit storage. Perpendicular recording will replace longitudinal recording in storage devices, placing bits on end instead of lying them parallel on the disc surface, thus dramatically increasing the possible storage density."
a crackdown on file-sharing. If they take that away from us, then whats the point of having that much space?
Legitimate content.
It's easy enough to end up with tens of thousands of photographs on your machine if you're in the habit of carrying a digital camera around. Now, think about what happens when you snap video clips the way you currently snap photographs.
This is already happening. With cameras being integrated into phones, it's growing even more.
Several reasons. Here are a couple. It takes less time for the heads to seek from the inside to the outsideof the platter. Smaller platters can be spun at higher rpms without flying apart.
With perpendicular recording the bar magnet would be standing on it's end.
Longitudianl recording is like this:
Perpendicular recording is like this: Google is your friend...Goodnight. :-]
- "Hear that?! The percolations are imminent! Cease your ingress!"
That's assuming current speeds. Well, as data gets more dense, the access speed inherently gets much faster, assuming the RPMs stay constant. If physical size stays the same, random access can't really get too much slower. So what is it that is going to be bad about terabyte disks?
The problem, as Jim Gray outlines it in the ACMQueue article:
There's a lot more about this in the article. Check it out; it's +5 Informative stuff.
Interesting? This clueless and sadly-late attepmpt at a FP is misleading everyone that reads it!. And you mods are to blame -- that's right: YOU!
::::::::
::::::::
Grrr, RTFA: there is nothing "3D" about it. It's still a 2-dimensional array of bits on a platter. The density increase comes from standing the little areas of magnetic media on end, instead of laying down. So, a top view of the old scheme would look like:
||||||||
||||||||
The new scheme, from the top:
In this case 2x density, as the lower one has twice as many dots in the same area as the dashes of the upper. (That is, each dot or dash represents the area of the physical medium used to store one bit by changing its magentic orientation). Get it? No 3-d. No holograms. Just 2-4x density increase by changing the orientation of the bits from parallel to perpendicular (relative to the disk platter surface).
everything in moderation
http://www.inphase-tech.com/technology/
You forgot 32-bit, 33 Mhz PCI bus @ ~150M/sec.
Faster hard drives would need a faster bus to operate off. I went looking for a non-server board the other day with PCI-X (for gigE), and couldn't find one in a store.
Drives aren't the only bottlenecks.
HDD manufacturers said they expect to start replacing 3.5in. disk drives with smaller 2.5in. devices in enterprise products sometime within the next year.
Why would they want to do this?
Average Access Time. Ever notice how it hasn't changed much in the last 20 years?
It was like 10-20ms in 1984 and is like 3-9ms now? No matter how fast you spin the disk or how much cache you add you still need to move the head from one side of the platter to the other. With 5" drives it was a little over 2" with 3.5" its a little over 1", with 2.5" drives 0.75" It's also true that if you make it smaller you can spin it faster, but I don't think 15,000 rpm is really hitting the limits of the materials or they would already have made the platers non-uniform in thickness. They could also go to single crystal metals like they do in aeroplane turbine blades (not so expensive to do in quantity.)
OTOH The disparity between bandwidth and access time is already embarrasing enough that I consider partitioning just half the space on my drives to improve access time. There are uses for big slow drives. For instance, things like audio and video if artists ever get their act together and jettison the media conglomerate dead weight they are carrying on their backs. Or for backups.
At this point GBs of hard drive space is like the Mhz thing was with processors. Most consumers just read the density and maybe the dBs and transfer rate, like they used to buy 900Mhz processors and get just 16 MBs of RAM when a 50Mhz Processor with 128MBs of RAM would have been literally thousands of times faster because they were thrashing with too little RAM. Buyers should look at access time, then transfer rate, and then capacity, unless it is for backups or some such tape replacement use. They should partition their drives because real-life filesystems still suck at placing frequently accessed data closely and contiguously for actual access patterns. If people realized this, hard drive manufacturers would do things like have multiple independent heads accessing the same platters, two would be easy, three could probably be done with current technology, and many more could be done with different mechanical linkages (for instance, screews might be slower and less elegant than an arm at moving the heads, but if you could fit fifty heads accessing the platters at once you would probably have better worst and average case access time.) This also would require updating some drivers, but I don't think it would take long considering the performance payoff.
If you have no idea what the difference between Longitudinal Recording and Perpendicular Recording might be, and the phrase "stands the bits on end" meant absolutely nothing to you because its an utterly ridiculous way to explain it, here's the lodown. Longitudinal recording is what we use today in everything from cassette tape to hard disks. It works by magnetising tiny sections of the recording medium. You can imagine the magnetised sections as tiny bar magnets laid end-to-end. The read head detects transitions in the direction of the magnetic field.
.
<- -> <-- -> <- -->
In the above diagram we're looking down at one track on the surface of a platter. Perpendicular recording works differently. The "magnets" or bits are arranged so that the field they emit is perpendicular to the medium, like this:
x . x . x . x
In the above diagram we're looking down at one track on the surface of a platter 'x' represents a field pointing away from us, '.' is one pointing towards us. This is what it looks like in cross-section (looking in from the edge of the platter):
^ | ^ | ^ | ^ | ^
| v | v | v | v |
In perpendicular recording the read head detects the actual direction of the fields emitted by these bits/magnets, rather than transitions in the field. Perpendicular recording is advantageous because it allows one to use a much smaller surface area on the medium for one bit. Imagine if you laid a line of bricks end-to-end on the ground, you could make the line shorter but taller if you stood each brick on end (so they're laid flat-to-flat), but you've not had to make the bricks any smaller in order to acheive this change in the length of your line.
Most of the above is hopefully right. Anyway it's a better explanation than that site gave.
Both schemes store the bit to some depth physically. You can't have an infinitely thin bit. Both schemes also still use a 2-D grid of bits. (Well, polar grid, since it's a spinning disc.)
A truly 3-D organization of bits within a single platter face would be something like those multi-layer DVDs, where within the same grid position you can access multiple bits by changing some aspect of the reading mechanism. (In the case of the DVDs, it's achieved by focusing the lense differently so only the desired layer is in-focus.)
--JoeProgram Intellivision!
ASCII art is great for porn but for technical stuff I prefer real images. This image cleared things up for me.
Talk about ignorant moderation. Sheesh.
Hard Drives technology is very mature. Every innovation has involved incremental improvments to the same basic tech. So the notion of Hard Drives getting more high tech is false. Second, the reliability of Hard Drives has been steadily increasing in a nearly linear fasion since their introduction in the 60s. There has always been instances of a particular Drive model or model family having difficulties. These are special cases from a statistical point of view. Saying that these models represent the quality of all Hard Drives is like saying that terrorists represents all Irishmen. On top of this, many HD reliability issues are realy HD handeling issues, i.e. originating with the PC manufacturers, not the HD producers. So the second part of the statement is also false, in fact way false.
how will this effect the reliability of future drives?
If you bothered to read the full artical, you would know that one of the hold ups of this new approch is quality concerns. The HD manufacturers will not deploy it untill it is suitable for their high end ( i.e. most reliable) Hard Drive lines.
Well, in my experience, engineers only like to brag about a new technique if it gives a 10x improvement. (Or more.) If you read the article, you would have noticed some numbers:
:)
:)
The "brick wall" in magnetic recording is called the superparamagnetic effect. This is the point at which the recorded data starts to get lost in the thermal noise of the media. (As you approach the superparamagnetic, it becomes statistically likely for recorded bits to sporadically flip states resulting in data corruption.)
For longitudinal recording technology, it is estimated that superparamagnetic will start to become a problem around 100Gbits/square inch recording density. (Current hard drive technology is around 50Gbits/square inch - so they are getting close to the wall.)
Perpendicular recording technology is estimate to scale up to around 1Tbit/square inch.
Now, what did I say about engineers liking to brag about 10x improvements? Well, 1Tbit is about 10x improvement over 100Gbit. How about that!
What this means to you: if current hard drives store about 120GB using a recording density of about 50Gbit/square inch, then we can expect perpendicular recording to eventually deliver drives that store about 2.4TB extrapolating up to a 1Tbit/square inch. Even if this technology only works half as good, at least we will eventually have hard drives that store 1TB!
On top of that, the article say they are moving away from 3.5" drives toward 2.5" even for "enterprise" applications. Now, if we get 1TB drives in 2.5" form factor that's going to result in some killer MP3^^err...uncompressed 24bit, 192kHz iPods
"faster (exponentially) solid state storage"
I know you won't (or can't), but please take a moment to learn what exponential actually means. It does not mean really big like many people treat it. We already have words for that like "really big". Exponential growth refers to growth whose rate of increase is proportional to its magnitude. That's all. It satisfies the equation x' = kx.
Solid state storage can be much faster than hard drive storage. There is nothing exponential about it. Incidentally there are various sorts of solid state storage. Some are faster than hard drive storage, some are not.
PlaY tried it. Remember them? They had a really neat technology - not bigger than CD but much, much smaller. It was self contained so you could toss a dozen in your pocket like coins. It was actually this close to being a killer technology, then they got too close to the RIAA and DRM'd themselves out of existence.
Hard drives are decent enough backup. They're now cheap enough to justify keeping a second drive just to duplicate everything on the first. But copying even 80Gb of data still takes damn forever, especially if the drives are in different boxes (I mean, if you're going to make a backup, you do want that backup protected in case of a power suply glitch... right?)
But a pocket full of sealed discs is a lot more convenient and error resistant than a case of CDs. Then again, the next generation commodity RAM is supposed to be magnetic, so maybe we'll finally get that convenient, portable storage in the form of actual solid state "coins!"
The "stacking bits on end" is done via two paths. First, head designs are being modified so that the magnetic fields are more vertically oriented. One of the buzzwords that will emerge is TMR for Tunneling Magneto-Resistive. The current dominant head design is GMR, or Giant Magneto-Resistive. For more details, do a Google search for both. The smaller the lateral width of the bit footprint, the higher the bit density. The other pathway to vertical recording is to change the shape of the crystals in the magnetic medium. As bit size decreases, it occupies only a few small primary crystals. If the crystals are made smaller, more uniform, and more vertically oriented, the bit footprint can be further reduced without compormising storage life. Relaxation of the magnetic field due to thermal noise, etc. is actually a much bigger problem, especially as the bit becomes distributed over fewer and fewer grains. Some of the current solutions to overcome stability problems involve stabilizing layers (e.g., Magic Dust). In any event, all the manufacturers are doing is to follow the HDD equivalent of Moore's law, which has held well since the 70s. Built it and it will get filled!
Although the solution proposed in the article would increase storage capacity by, say, a factor 2 or 4, it still is a temporary solution that does not solve the fundamental problem at hand.
The fundamental problem is the superparamagnetic limit: if you make a magnetic domain (a bit) smaller than a certain size, it becomes thermodynamically unstable. In English, this means that very small bits loose their value after a while. It also means that for the time being, we'll have to use tricks to pack the bits closer together while keeping them large enough to be stable.
It should be noted that perpendicular recording is not the only effort to achieve higher recording densities in the looming shadow of the superparamagnetic limit. Indeed, harddrive manufacturers have seen this problem coming for a number of years now, and have had meeting to discuss possible solutions.
On a brighter note, there seems to be progress in circumventing the superparamagnetic limit: very recent research show promising results for the future.
Support a Europe-related section on Slashdot!
Yes, sequential data transfer is increasing up to 40 and 50MB/s, but the "speed" of the drive in terms of RPM is what affects seek time. Most data access for average users involves getting data from all over the disk and the vast majority of the time spent is wasted moving the little heads from track to track.
10,000rpm Serial ATA drives are now available. I have a couple in a machine at work, but they are LOUD, and still fairly expensive.
Um, I meant *exactly* what I said. I'd like to see access times of new drives be fractions of what they are today, and transfer speeds exponentially, not incrementally, higher.
The point is that "exponentially higher" doesn't mean much if the exponent is very small, or if the duration of a large exponent is very small. That being the case, "exponentially higher" doesn't really mean anything unless you quantify it in terms of rate and time, or at least specify some existing linear growth rate whose coefficient you'd prefer to be used as an exponent.
Saying "I'd like to see transfer speeds get 10 times as fast every year for a few years" would be a meaningful statement. Saying "I'd like to see transfer speeds get 1% faster every decade for a few decades" is also a meaningful statement, and also describes exponential growth, but probably isn't what you wanted to say. Saying "I'd like to see transfer speeds 100,000 times faster than what they are now" is also meaningful, and may be what you want, but it isn't necessarily exponential.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
Historical Note: This "new" recording technique is the same one that failed to take hold in 2.88M EHD floppy disks about 15 years ago. Back then the new recording material was barium ferrite, whose magnetic domains arrange themselves vertically with respect to the substrate.
Compared to ordinary floppy disks with horizontal magnetic domains, this technology had the potential of increasing data densities by as much as 2-3 orders of magnitude. Unfortunately, the new disks were expensive and not compatible with the huge installed base of 1.44M drives. EHD drives required BIOS changes that weren't possible in those non-FlashBIOS days. Even if those problems could have been solved, IOMEGA's Zip drives were offering far more bang for the buck.
Of course, none of this would matter for hard drives.
It'd be more like:
from --------
to ||||||||
Perpendicular recording would effectively replace a two-dimensional bit with a one-dimensional bit, from the recording head's point of view. (Or something close to that.)
I'd have a personalized plate on my car, but "toxic bachelor" won't fit into 7 letters.
The rotational speed of the drive is directly related to the access time. If the data you want is on the other side of the platter, you must wait for it to rotate 180 degrees before you can start reading, regardless of whether the disc is 1/2" in diameter or 2" diameter, whether there are 1GB per square inch or 10GB per square inch.
When the head gets lined up with the track and ready to read, the data it's waiting for can be anywhere between 0 degrees and 360 degrees away. If you average out all those possibilities, you can expect the data to be about 180 degrees away.
Now, a 15000 rpm drive rotates 180 degrees 30000 times per minute. Conversely, it takes 2ms to rotate 180 degrees. If you consider that a typical 15k rpm drive has an average seek time of 3.3ms and we know that 2ms are spent waiting for the disk to spin, than 1.3ms must be spent moving the head. This proves to me that rotational speed is more important to access time than data density.
I'm no hard drive engineer, but I would bet that an increase in density would mean a decrease for rotational speed since a read head probably has a limited bandwidth. (This is probably why the faster-spinning drives typically hold less data.) If you halve the time moving the head while doubling the time waiting for spinning data, you will see an overall increase in seek time.
My conclusion is that greater density and less rpms would hurt access time which is the most important performance factor. However, like the "MHz myth", I'm sure marketing will focus on bandwidth benchmarks for performance instead of real-life application performance.
So, a top view of the old scheme would look like:
::::::::
::::::::
..1....0....1....1....0..
||||||||
||||||||
The new scheme, from the top:
In this case 2x density, as the lower one has twice as many dots in the same area as the dashes of the upper.
I don't think that's quite right.
Unless I missed a transition from longitudinal to transverse recording, the old scheme produced a track like this (viewed either from the top or side:)
N---SS---NN--------SS---N
The vertical scheme lays the magnets INTO the medium rather than ALONG the track. Viewed from the side:
NSNNS
|||||
SNSSN
10110
Or, viewed from the top:
NSNNS
The problem with longitudinal recording is that, as you make the bits shorter, the magnetic fields of adjacent opposite-sense bits become more effective at trying to flip the singleton to go along with them. (Magnetic domains are more self-reenforcing, and thus stable, when they're long and thin, subject to flipping from thermal agitation at progressively lower temperatures as they become more short and fat.)
Make the bit too short and the neighbors "squeeze it out":
N--SSNN--S -> N--------S
1.1.0.1.1. -> 1.1.1.1.1.
But with vertical recording the adjacent, opposite-sense neighbors tend to STABILIZE the bit, and the smaller it gets, the more stable it gets. (And you're guaranteed a limit on the number of long runs of same-sense by the coding scheme, which has to flip now and then to keep the read electronics in sync.)
So you can shrink it WAY down - both along the track and across it - to the limit of the head technology to produce the original magnitizing field or the inherent domain size of the magnetic medium.
You can get FAR more than a factor of two in EACH direction - and multiply the two improvements to get the increase in bits per unit area.
(They've been talking about this for years. How come it's just hitting the field now? Did they go to transverse magnetization in the meantime? That would have similar advantages of smaller-is-more-stable. But the track would be far wider than with vertical, as would the gap, so you'd still save a bunch in one of the dimensions by standing the magnets on their head and packing them in tightly, like a bundle of sticks, rather than laying them on their sides.)
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way