Hacker News new | past | comments | ask | show | jobs | submit login
Magnetic recording breakthrough reported (sciencedaily.com)
101 points by arjn on Feb 8, 2012 | hide | past | web | favorite | 32 comments

Update: I tried to find some more info but unfortunately it looks like the original paper is behind a pay-wall. Here are some related links : http://www.psi.ch/media/using-heat-for-storing-data


Since this doesn't affect read speed, it's interesting to imagine what would happen if there were a de-facto asymmetry between read and write speed for hard disks. I'd guess you might end up in a world where a "disk" actually becomes a "RAID-array-in-a-box" so parallel reads could make the read speed be close to the write speed.

If the heat pulses only toggle the state of the nano-islands, then it makes little practical sense to speak of 'write speed,' since you'll have to determine the current state of the bits (that is, you'll have to read them) before you attempt to write them.

Terabytes per second? Really?? Given that SATA-3 is still at 6Gbps, this would be 1300x faster than SATA-3. How would you get data to a drive that fast for it to write?

Sort of a moot point regardless. If you have a spinning disk the average latency and maximum read or write speed is determined by the speed of the disk.

Not necessarily. For example you could have many read heads and read from the disk much faster by reading in parallel. It's complicated to do, but it is possible.

You can also increase the areal density, which means more bits fly past the head per second, which also increases speed, even though you can't increase the rotational or seek speed.

I would expect that's how they plan on getting to those ridiculous write speeds. Decrease the size of 1 bit so that more of them would fit on one track (a circle on a platter). They would be able to do that since the time to flip the region of the bit would be smaller. However, there is still the problem of finding a material with that much smaller magnetic regions. Which is already a problem in current drives (perpendicular recording) so I don't expect this to fly.

Increasing the number of platters improves read/write throughput independent of the low-level technology used. As far as increasing density, that only scales relative to the square root of areal density. If you increase density by a factor of 100 you only increase the number of bits per track by a factor of 10.

The same rules don't apply to mram, of course, but from the article I'm not entirely sure if these techniques are useful for such designs.

I tend to believe that in the short/mid future consumer hardware will mostly consist of SSD or similiar solid state forms of memory. Abandoning spinning disks and relying on much faster, smaller and more reliable memory seems so much more promising than hard disks.

Yes. But remember, tape was displaced from the mainstream, but lives on in professional backups. There will still be lots of profitable applications for spinning disk.

I'm confused. Doesn't a writable DVD work on the same principle of using a laser to heat small portions of a spinning disc? Why should this methods be so many orders of magnitude faster than writing to a DVD?

First, keep in mind that this is a modification to hard drives and not DVD-like equipment. The heat pulses in the case of the article are still modifying the magnetic bits on a spinning metal disk.

Second, it seems that it's faster to generate these heat pulses than it is to switch the magnetic orientation of the write head. I haven't looked into the physics, but this is the only plausible explanation my limited imagination can fathom.

Does this affect read speed at all? The article only mentions write speed.

The readout will still be carried out by a magnetoresistive sensor, so that is not directly affected. With a smaller bit size it might be necessary to further develop the read head, but the road map for that is quite well understood at the moment.

Didn't minidiscs read the data optically? Could that tech be combined with this to make a pure-laser hard disk?

Can someone smarter at physics than I am explain how this works?

The article specifically suggests use in hard drives, but how soon could this realistically be used in commercial hard drives?

I'm not a physicist, this is what i understand : They've found that very quick heating of a magnetic dipole can cause it to flip. Currently this is flipping is done by a magnetic head (in hard disk drives). Since bursts of infra red light can be used for heat pulses, recording can be sped up considerably.

How could that ever work, for infrared to affect magnetism?

You ignored the article and what the GP said: Through heat. Not (directly) by the light.

In response to your question on availability in commercial HDs, I think that unless a comparable significant advance is made in read speeds, I doubt that you will find it in consumer grade HDDs soon. A disk that can read at e.g. 100mb/s and write at 50gb/s would be not easy to sell to the consumer at a premium price (where do I get data faster than my HD/SSD read speeds from to store?).

I assume though, that it might be used in e.g. CERN, where they need to store a ridiculous amount of data in very short time. Or similar activities where the write speeds are the limiting factor.

I could definitely use great write speed and average read speed.

Use RAID-1 (mirroring) and have lots of disks in parallel. You can combine them for great read speed, but you still have to write to every one of them so writing is not improved.

If your write speed was much faster than read then this would work perfectly.

It would also find applications in things like data archiving, warehousing. Or buffering huge amounts of data for later processing (like the large hadron collider).

When I was at FNAL on the USCMS data team, we were able to take data from the LHC at ~40Gbps over an optical link and write it out just fine to spinning disk (which would hold it until we could shove it out onto thousands of LTO4 tapes).

I don't think there are going to be a ton of applications for this hardware, but it'll be great for those edge cases where you absolutely need that sort of data rate.

my question being, wouldn't these drives be much more susceptible to rapid and massive data loss from changes in ambient or operating temperature?

not any more than regular HDDs. The method described will probably be applied to existing HDD platter technology (my opinon)

How would the performance compare to a SSD?

Well the article mentions that:

"This revolutionary method allows the recording of Terabytes (thousands of Gigabytes) of information per second"

Therefore, If you can write 1TB/sec the seek it would pretty much outperform any SSD cluster, in terms of writespeed. unfortunately, nothing is said about read speeds/access times and I assume that that will be a harder problem. Because they are still storing the data magnetically and the platters don't emit heat based on their polarisation, they still have to pretty much read like normal hdds do. You cannot (until now, like it was not possible to alter the magnetic polarisation of metals other than with another magnetic field) sense the polarisation except for with another magnetic field.

I am really interested in how that will turn out, but before they make a significant advance in read speeds I doubt that the technology will make it into consumer grade HDD's any time soon.

SSD read/write speed is pretty much limited by cost and bus speeds, not technology, as for the foreseeable future you can boost it by putting more chips in parallel (which is why current high end SSD solutions are pretty much all on PCIe cards, because they're already way too fast for SAS or SATA etc.). The same would probably become the limiting factor for this technology too, even if they can solve the read speed issue.

On top of the read problem, they have to deal with the fact that HDDs are just really energy-inefficient. With the world going to mobile, we can't predicate the future on the need for moving parts.

It's a damned cool technology, and I hope it leads to something useful, but I just can't see it becoming the prevalent paradigm.

Yes, you're right the world is going mobile and in mobile less (no) moving parts is better (more durable etc.) That is also why I think that consumer grade devices will be unlikely (unless read speeds are bumped up, too)

But mobile, in our case, means also a shift to the cloud, where significant more potent storage technologies could make a real difference. I really hope, too, that technology will lead to something new.

Fair enough. It's hard to see where these kinds of write (but normal read) speeds would be necessary, even in a server farm, but I guess the beauty of enabling technologies is that they allow for use cases no one could have imagined when they were conceived.

Shooting from the hip here; I expect seek disparity would remain about the same, while maximum bandwidth would skyrocket.

Actually, hold on, I wonder if seek would improve very noticeably? Is seek time mostly the long-distance movements of the head, or the locating of the file in-track after the head has traveled? If it is the latter, you might be able to boost seek quite a lot.

"Seek time" as you are putting it is almost all the lateral movement of the read head (not the rotation).

Though what you're really going for here is "access time", which is "seek time" (properly defined as just the lateral movement of the head) + "rotation time" (what you call "locating of the file", which is actually just sitting there waiting for your 7,2k rpm to come around once every 4ms or so - rotating disks rotate at a constant rate, you can just do the math). A Western Digital Caviar Blue (random benchmark, I only know this because it was the last drive I bought) has a ~9ms seek time.

To recap:

Access time = seek time + rotation time ( + negligible other times, well under 10~100us)

Rotation time ~4ms (3ms for 10krpm) Seek time ~9ms (supposedly ~4 for really high end drives)

Now this is just with the technology we have right now. I do not know enough about the constraints on hardware to know if they could push the limits. For instance, if the actual write (/read?) bandwidth could theoretically be higher, would it be possible to just rotate the disk faster and/or bump up the head seek time?

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact