5.6TB helium disks could balloon, lift WD onto enterprise throne Western Digital could topple Seagate as the king of enterprise disk shipments thanks to its helium drives, according to analysts IHS. The technology is being developed by WD's subsidiary Hitachi GST. Read-write heads and spinning magnetic platters in hard drives filled with helium experience less drag rather those in devices …
Ever seen those foil balloons? There is a reason they're not made out of rubber. Helium molecules are so small they pass easily through rubber. Even the foil balloons can't retain the helium indefinitely. So it would be interesting to hear how they solve this problem with these drives. The cynic in me says sees "planned Obsolescence.
Helium is SO tiny (it has an atomic weight of just four and, being a noble gas, typically exists as isolated atoms) it can pass through even the tiniest of gaps. This includes nanoscopic holes in otherwise-solid plastic and metal. The casings need to be carefully treated to render them truly gas-tight.
With similar pressure on the inside and outside of the drive, leakage should be pretty low. In fact, equalising the pressure inside and outside could be the real headache here. Currently, drives have a small vent and various filters. Some times, a long tube ingeniously created by embossing a pattern on the outer casing too. If you've got to keep the same gas inside for years, there'll need to be a new system, though I don't know what.
Alternatively, maybe they are completely sealed. Start just above atmospheric pressure, and the tiny bit of leakage outwards shouldn't be much of an issue. Air getting in would be a lot worse, but if the pressure is always a bit positive, it shouldn't happen much.
There you go Seagate, that's how you do it. Royalties please.
They fill the electrical generators at power stations up with hydrogen which acts as a coolant, and reduces drag. This has been done for decades. 'Course there they have a slight over-pressure, a shed-load of monitoring, and decent extraction systems to prevent a bang, but it goes to show that with a bit of thought and design effort, hydrogen could be a viable alternative.
Thing is, you don't have a lot of them (compared to say millions of hard drives) and being a generating station you would have trained personnel on hand should the hydrogen content approach the danger zone (that's the thing--as long as the atmosphere in there is too rich, there isn't enough oxygen to allow combustion). Because once it does get into the danger zone, you're looking at a bomb ready to blow any moment.
These checks against safety aren't possible in a consumer environment. Besides, given our litigious nature, the first exploding hard drive will likely result in a swarm of lawsuits.
If you already make your hard drive enclosure air tite so it can be permanently gas filled, why not go to deep vacuum instead? Then there is no aerodynamic drag and an automatic indication of warranty failure for the enclosure.
Helium is relatively expensive and getting more so, vacuum is cheap. Certainly there will be FAR LESS molecules in contact with the drive platters. Heat removal would have to be entirely by direct conduction to the enclosure.
However, you can't beat the cost of "Nothing".
Ahh, the power of the naysayer, able to defeat/prevent great ideas before anyone even begins to think them through seriously.
Apparently, you didn't read the bit I wrote about having to remove heat by conduction? I have to be more concise to avoid repeated downvotes from those who can't think outsdie the box.
Next, you can't leave a breathing hole in the design IF you use rare gas or vacuum, the person mentioning
"Then change the design" was the most rational response including the comments on the read/write head design requirements.
Why does the head need to "float" on anything at all? It would appear to me at least that a rigid arm that had the right clearance above the platter is all that you need. Using the Bernoulli effect (aerodynamic lift) for spacing the head from the platter was probably done in convenience only to accomodate the issues of an air/gas filled drive enclosure.
The vacuum design would make for zero turbulence at the R/W head and I'd assume it would be easier to keep stable and have faster angular movement too.
As far as needing "bearings" there is such a thing as non contact permanent magnet bearings (that could be shielded by MU metal to prevent data loss or corrruption) that could allow spin rates even higher than 15,000 rpm.
Correct me if I am wrong, but wouldn't the platter MOTOR will produce the most heat, followed by conventional bearings and the head drive motor. ALL could be in direct contact with the outer case of the drive so heat dissipation would not be so difficult to achieve.
Why does the head need to "float" on anything at all? It would appear to me at least that a rigid arm that had the right clearance above the platter is all that you need. Using the Bernoulli effect (aerodynamic lift) for spacing the head from the platter was probably done in convenience only to accomodate the issues of an air/gas filled drive enclosure.
Because the tolerances at the microscopic sizes of the hard drive heads are practically beyond pure human engineering to reach on a mass scale. The Bernoulli Effect provides a much-needed margin of error by ensuring the heads stay a safe distance away from the platter. Without that cushion, given that both the platter and especially the head can undergo flexes or heat expansions that can cause them to come into contact (and since the arms are so small and so thin and the gap so tiny, it wouldn't take much)..
There's no way you could maintain a precisely defined tiny gap between the head on the end of a two-inch arm and the spinning disk in a vacuum. The head quite literally flies. It's the dynamics if the gas between the head and the disk that keeps it at the constanrt correct distance.
BTW what about Methane? It's about two-thirds the molecular weight of air (CH4 = 20, O2 = 32, N2 = 28), it's not prone to leaking through nano-pores like Helium is, it's not prone to reacting with metals like Hydrogen is, and although it's flammable it's hard to see how an HDA-full of methane could cause any sort of serious problem. Lighting a gas cooker burner probably lets more than an HDA-full out into the air and then deliberately sparks it.
By way of ullustration: If a HDD head was the size of a jumbo, it'd be "flying" at an altitude of ~50mm. On the same scale, a fingerprint is about the same height as a Mini. Or so I was told once, by a chap from the HDD industry.
That were then, when I were a lad. The clearances are probably tighter now. On the same scale, the icon is a head crash.
Yup it's called the Bernoulli effect. It has something to do with a moving fluid stream and pressure differentials.
The spinning disk drags air/helium with it so the head is stationary in a moving airstream, the head acts like a wing and is lifted off the platter.
Helium is less dense than the Nitrogen/Oxygen mixture that is air so drag is reduced and less power is required to rotate the disk at the same speed.
Vacuum has at least four disadvantages for hard disk enclosure contents...
1. It is an extremely effective thermal insulator. Any gas or liquid can carry heat by convection, conduction, and radiation. Vacuum only carries heat by radiation, which will make cooling internal components interesting.
2. Gas is required by all hard disk technologies since the dawn of time to support the heads, so that they "fly" an extremely small distance away from the platter surface. Vacuum doesn't allow this.
3. Your vacuum-insulated non-flying hard disk also has to withstand 15psi of crushing forces operating continuously.
4. All components inside the enclosure have to be vacuum-tolerant. Since bearings are notoriously hard to make vacuum-tight, the motor needs to be inside the vacuum enclosure, and all the bearings must therefore be lubrication-free or use vacuum-tolerant lubricants. Many plastics are hard to use in a vacuum, as the additives that make them flexible are slightly volatile and will out-gas into a vacuum. Exposed metal surfaces must be protected against scratching so that you don't get metal-to-metal vacuum welds spreading tiny conductive fragments around the interior of the enclosure.
The list goes on and on, but the overall conclusion for me is that vacuum-filled hard disks are a non-starter, and even if it is technologically feasible, the additional cost would be prohibitive.
or in this case helium bearing for the heads to float there is no current way to have such a close tolerance without the use of a gaseous medium to "float" the heads on
I wondered about the whole "helium is a bugger to keep hold of" thing and not only that what will all the MRI's do? not more waste of that precious helium... why don't we just put the stuff in balloons or something sheesh!
that said a partial vacuum or lower pressure might work just as well would probably need to spin the platters a bit faster though for the same effect to happen
<----------- the one with the vacuum flask of scotch in the pocket
The now ancient slim 5.25" Quantum BigFoot hard disk drives it's the design that manufacturers should pursue now for highest capacity drives.
Yes, it would be limited to full desktop and server cases BUT it would allow 8TB to 16TB or more with current technology and 2 to 4 platters.
Helium filled drives in a 5.25" slim design would allow for very high capacity disks to happen now at current manufacturing costs and not in 3 or 5 years.
Unfortunately, raw storage isn't the only limitation. You also need to be able to effectively seek and read/write the tracks. Increasing the diameter of the medium increases tracks, but also decreases IOPS per TB as you have N heads accessing significantly more physical space. You can pull tricks like putting multiple heads per platter, but they run into limitations also.
I'm also not convinced current technology could support a 5.25" platter at 7200 or 10k RPM - the centrifugal forces at the edges will be significantly larger than at 3.5". The Quantum BigFoot drives didn't spin anywhere close to 7200RPM AFAIK.
@Joerg
Increasing data capcity per-drive is already infeasible if your data is important enough to keep (and if it's not, don't record it in the first place).
Specifically, modern drives (e.g. all currently available 4TB drives) come with a manufacturer spec of the error rate of 10^-14. That means one unrecoverable bit (after all ECC is applied, which means you lose at least the whole physical sector) evern 10^14 bits, which is roughly every 11TB. That means that if you have a mirrored pair of disks and one of them fails, the chances of encountering an unrecoverable error during resilvering is approximately 36%. Figures get worse for 3-disk RAID5, and to maintain a reasonable chance you can still get your data back intact, you really need to be using 4-disk RAID6 with 4TB disks, which is already silly because suddenly you need 2-3x as many disks as you have data to keep a reasonable level of redundancy.
The short term solution is to switch to smaller drives, both physically and logically. Switch to 2.5" disks and the capacity comes down to 1TB/disk, and error rate and reliability being the same, you are in a much saner area of the risk curve.
The only long term solution is to improve reliability of disks in line with capacity, but since reliability (10^-14 error rate was fantastic 20 years ago, but is next to useless today) and performance of disks have remained static (7200rpm = 120 IOPS and everything else is just re-arranging the deck chairs on the titanic), it is very difficult to envisage a large leap in either arriving any time soon, at least as far as spinning rust is concerned.
Current discs offer 100MB/second read rates, so you're saying that if you constantly read over a disc you'll get an unrecoverable error every other day.
This doesn't seem consonant with something like http://www.numberworld.org/misc_runs/pi-10t/details.html ; yes, this lost a lot of time to disc failures, but in an environment where it was running flat-out to 24 separate spindles without redundancy it lost one disc about every four spindle-years.
@Tom Womack:
Essentially - yes. If you read over and over the disk, you will eventually find errors. How do you think pending and reallocated sectors come about on the disk? Eventually a sector becomes unreadable. If you are sequentially reading the disk you probably have a slightly lower probability of this occurring than in a more realistic random seek case, but it will happen. I run a full array scrub (ZFS) plus a long SMART self-test which does the same thing, but sequentially once per week with my 1TB drives, and statistically that would mean I get one reallocated sector every 6 weeks (assuming no other load).
Randomly looking at one of the older disks in one of my arrays, I see that it has 48 reallocated sectors over 34369 power-on hours or approximately 204 weeks. If I was getting an unrecoverable sector every 6 weeks I would expect roughly 34 reallocated sectors to be showing up. I see 48 reallocated sectors which seems roughly right for the sort of load that particular array is under. So I would say the 10^-14 error rate seems pretty much spot on in this case, at least for various Seagate 1TB drives I'm using.
Note that the error rates are NOT the same as failure rates. Failure is generally deemed a complete failure where the disk doesn't work at all (for example, but not limited to, it runs out of spare sectors for reallocations). Failure rates are, in my experience at least, very model specific (e.g. ST31000340AS disks I see failing all over the place, they are averaging more than one replacement per disk under warranty, but ST31000333AS have been quite good, only had a handful die, whereas 524AS/528AS drives I've not seen any complete failures on, but I've only been getting those recently in comparison).
Anyway, the point is that sectors going bad still means data loss, which you can only avoid using RAID or recover from using a backup. Most people would consider any data loss to be a bad thing, because that sector could be something easily replacable, e.g. an OS file, or it could be something completely irreplacable (e.g. a segment of family photo of which you have no other backups). Way too few people stop to consider such things these days.
Since when RAID-5 had an higher failure rate than single drives?
Also no one with a bit of brain would setup a 3-disk RAID-5 anyway. The minimum proper RAID-5 setup is using 4 drives.
Also.. telling that current drives would have 1 error every 10E14 it's just not true.
Enterprise SATA and SAS hard drives have 10E15, 10E16 and 10E17 error rates nowadays.
If you thought that consumer cheapest hard drives with 10E14 was the best on the market today then you were living in the past.
@Joerg:
You clearly didn't fully understand what I was talking about. I never implied that RAIDn (n>0) has higher failure rates than a single drive. What I was explaining is that EVEN WITH RAIDn (n>0) the chances of losing data (albeit probably not ALL data on the array) is very much non-trivial.
As for RAID5 and the number of disks - you clearly do not understand how RAID works and how you size the number of data disks vs. the number of parity disks in terms of redundancy required to maintain a particular degree of data safety for any particular set of disks.
For example, if you have a 4TB disks with 10^-14 error rate, and have them in a 3-disk RAID5, (8TB usable) that gives you a 72% probability of losing a sector's worth of data following a single disk failure, during resilvering onto a fresh, new disk.
If you were to use 1TB disks of same reliability, to achieve the same 72% chance of losing a sector during resilvering a failed disk, you could use 9x1TB disks (8 data bearing, +1 parity disk).
RAID5 (or any other RAID level for that matter) has no magical "optimum" in terms of using at least 4 disks for a "proper" setup. It is all worked out as a function relating the size of the drive, the error rate of the drive, and the number of disks in the array with the required probability of not losing any data in case of a disk failure.
There are no mechanical disks on the market that I am aware of with the manufacturer specified error rate of less than 10^-15 (e.g. Seagate Constellation is marketed as an enterprise grade disk, and has 10^-15 error rate specified). And yes, of course, this helps a great deal because it pushes the volume of reads per bad sector from 1 per 11TB to 1 per 110TB. But most people use vanilla desktop grade drives without actually considering the chances of losing some of their data.
Anyway, the original point was that pushing the capacity of a single disk further without a corresponding reduction in unrecoverable error rates would be quite dangerous.
@AC:
As explained above, mirroring is no longer a sensible option with 4TB drives and 10^-14 unrecoverable error rate. Except maybe if you use at least triple mirroring. I'm sticking with 4-disk RAID6 + backups with 4TB drives for now.
In most hard drives, there's an itty bitty hole to allow the hard drives internal pressure to equalise with the external air pressure. I'm guessing that drives using helium would have to necessarily be sealed and pressurised, Which could cause problems for the drive heads when the helium eventually leaks out.
SSD won't replace mechancial Hard Disk drives for another 10 to 15 years at least.
Manufacturers already have technologies in lab to increase Hard Disk capacity over 300TB. It's just for a marketing reason that they aren't pushing the technology and are slowly re-using existing tech before releasing anything new.
quote: "Manufacturers already have technologies in lab to increase Hard Disk capacity over 300TB"
Is that for 300TB spinning platter HD, or 300TB SSD? Do they have a bus that can support better data transfer rates that mean it won't take a year to write that 300TB?
Are they also holding back on cold fusion and 90+% efficient solar cells?
It's not that I don't believe corporations are capable of that kind of subterfuge-for-profit (hi there ebook pricing!), it's that too many people would end up knowing about it for them to keep it particularly secret; humans are normally the weakest link in any security/secrecy chain. Yet your post is the first I've heard of >10TB capacity in a single enclosure.
Colour me sceptical...
http://www.itwire.com/your-it-news/entertainment/8350-seagate-to-offer-300-tb-hard-drive-by-2010
" Seagate to offer 300 TB hard drive by 2010 03 January 2007 By Alex Zaharov-Reutt
-Correction- The 300 TB is actually terabits, and not terabytes. Therefore, the new Seagate drive in 2010 will store approximately 37.5 terabytes"
http://www.bit-tech.net/news/hardware/2007/01/10/seagate_says_300tb_aint_happening_in_2010/1
"Seagate says: 300Tb ain't happening in 2010 Published on 10th January 2007 by Tim Smalley Seagate research's estimates are that 50 terabit-per-square-inch density may be achievable using HAMR with perhaps a combination of Bit Patterned Media but that's moving well past the 2010 timeframe"
The 100mph carburettor at least (or rather, 100mpg fuel injection car) is meant to be here inside the next few years. Perhaps the others will follow.
http://www.autocar.co.uk/car-news/new-cars/vw-aims-100mpg-golf
In any event, the 300TB HDDs will just exacerbate the fundamental problem that capacity goes up to the square of disk area whilst sequential access only goes up linearly and random access barely improves at all. Simply the larger the HDD the worse the I/O bottleneck becomes as data rates increase at a much faster rate than the ability to access the data. It doesn't much matter if the drive is filled with air or helium You simply can't spin the platters any faster as they are pulled out of shape with the forces involved, the drives become unstable and bearings give out.
Having 300Tb in the lab is not the same as it being reliable, or economic to manufacture/sell it..
1Tb SSDs are already available - if you can afford them. At current rates they'll be affordable in about 18 months.
The big issue is that people try to put big storage and fast storage in the same drive enclosure and that ends up being costlier and kludgier than using separate solutions. There are better ways.
Just have more Alpha decay!
Need more Alpha decay, more nuclear power plants. Nice side effect.
Then again, you could have fusion plants, but that is a ways away.
Yes, Hydrogen is a bit nasty. It makes metals brittle, and corrodes quite nicely. Best to make it on the spot, a little Zinc in HCl works quite nicely for this (as I saw in 7th grade/First Form).
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
