Helium filled 10TB harddrive

[Srbel]

As far as I know, MTBF means nothing. They might as well make up the numbers, if they don't do already.

You are correct. In general, MTBF is complete frogshit for an entity like a hard disk.

What is important is MTTF.

http://www.weibull.com/hotwire/issue94/relbasics94.htm

MTBF for hard disk is equivalent to MTTF since a hard disk is not repairable. You could more accurately use MTBF on the computer system that the hard disk is installed in, but that is another discussion. In any case, MTBF is not a nonsense term but you need to understand what it means. It is the mean time between failures assuming that all disks are replaced at EOL (essentially end of warranty). MTBF of 2.5M hours does not mean that the disk is expected to operate for 2.5 million hours (285 years) before failing. It means more something like if you have 1000 such disks running, then you will see one failure per 2500 hours, or 104 days. After the lifetime of the disks is used up (~5 years), you need to replace them all, otherwise the failure rate will be expected to shoot up.  Really, the industry should use the term MTBF_SR (MTBF with schedule replacements). The internet archive seems to keep good records their hard disk failures and publishes lots of data about them for those interested.

Than if I have 10 disks running, I will see one failure per 250.000 hours, or 28,5 years?

Bollocks...

[rs20]

Than if I have 10 disks running, I will see one failure per 250.000 hours, or 28,5 years?

If you preemptively replace the disks every 5 years, keep them in a cool, shock-free data center, etc, etc -- then yes.

[fivefish]

Helium filled hard drives.... that's going to have a fast data transfer rate, it will make MP3s sound like Alvin and the chipmunks. 

[wraper]

In comparison, in the recent 5 years, I own and owned at least 14 SSDs, and none of them failed till the point I sold them. The oldest one I'm current using is 2 years old, with its temperature constantly at 50C (in NUC, its SSD bay is hotter than its HDD bay), not even a single SMART warning.

My point is HDD will eventually die out. During black friday, the cheapest 960G SSD goes down to $150-ish. $800/10T=$0.08/G, $150/1T=$0.15/G. Not too much cost advantage to me.

Most SSDs don't have temperature sensor and just report some fake temperature.

[suicidaleggroll]

Most SSDs don't have temperature sensor and just report some fake temperature.

All of mine do...Samsung, Intel, and Sandisk.

[kc8apf]

As previously noted, the drive is a normal cast aluminum base with a laser welded sheet over it.  The helium (actually a mixture of various gases that they won't reveal but mostly helium) is at a very slight positive pressure but the total amount is really small.  The primary reason for the helium is to reduce turbulence when flying the heads over the platters.  Reduced turbulence means the heads can safely fly closer to the platters which means fitting 7 slightly thinner platters in the same physical envelope.

Regarding the amount of helium, when HGST first started talking to my team at work about these drives over 2 years ago, I quickly pointed out that we crush drives on a near continuous basis as part of our data security program.  They did a report for us to show that crushing 1000+ drives a day in a 20'x20'x10' room would be perfectly safe and no one would have high-pitched voices at the end of a shift.

[macboy]

As far as I know, MTBF means nothing. They might as well make up the numbers, if they don't do already.

You are correct. In general, MTBF is complete frogshit for an entity like a hard disk.

What is important is MTTF.

http://www.weibull.com/hotwire/issue94/relbasics94.htm

MTBF for hard disk is equivalent to MTTF since a hard disk is not repairable. You could more accurately use MTBF on the computer system that the hard disk is installed in, but that is another discussion. In any case, MTBF is not a nonsense term but you need to understand what it means. It is the mean time between failures assuming that all disks are replaced at EOL (essentially end of warranty). MTBF of 2.5M hours does not mean that the disk is expected to operate for 2.5 million hours (285 years) before failing. It means more something like if you have 1000 such disks running, then you will see one failure per 2500 hours, or 104 days. After the lifetime of the disks is used up (~5 years), you need to replace them all, otherwise the failure rate will be expected to shoot up.  Really, the industry should use the term MTBF_SR (MTBF with schedule replacements). The internet archive seems to keep good records their hard disk failures and publishes lots of data about them for those interested.

Than if I have 10 disks running, I will see one failure per 250.000 hours, or 28,5 years?

Bollocks...

Absolutely, if you replace them when their warranty is up, regardless of whether they appear to be working fine. For cheap consumer disks, this means every 1 to 2 years. For expensive enterprise disks, once every 5 years (or so).

[David Spicer]

If by 2.5M hours they mean 2.5 million, that's close to 300 years.
I didn't kno they had Hard disks in 1715,
Make you wonder. 
Or maybe they tested 300 disks for a year, I guess.

[rs20]

If by 2.5M hours they mean 2.5 million, that's close to 300 years.
I didn't kno they had Hard disks in 1715,
Make you wonder. 
Or maybe they tested 300 disks for a year, I guess.

I heartily encourage you to read the rest of this thread 

[Circlotron]

Apparently a layer of graphene a single atom thick is completely impermeable to helium, so it's not impossible.

Would make a neat party balloon.

[Rick Law]

As previously noted, the drive is a normal cast aluminum base with a laser welded sheet over it.  The helium (actually a mixture of various gases that they won't reveal but mostly helium) is at a very slight positive pressure but the total amount is really small.  The primary reason for the helium is to reduce turbulence when flying the heads over the platters.  Reduced turbulence means the heads can safely fly closer to the platters which means fitting 7 slightly thinner platters in the same physical envelope.

Regarding the amount of helium, when HGST first started talking to my team at work about these drives over 2 years ago, I quickly pointed out that we crush drives on a near continuous basis as part of our data security program.  They did a report for us to show that crushing 1000+ drives a day in a 20'x20'x10' room would be perfectly safe and no one would have high-pitched voices at the end of a shift.

[bold added to quote]

Interesting!  I was guessing it was because helium is a noble gas and would not react with anything thereby eliminating oxidation and other degeneration induced by chemical reactions.

[T3sl4co1l]

Helium is usually chosen for thermal conductivity, e.g. in turbines (Brayton cycle?), bearings, etc.

I don't know that the fluid properties, such as turbulence, are significantly different (i.e., the flow regime at the surface of a disk spinning that fast, will still be largely turbulent).  I doubt it's eliminated, but reduced is possible.

If chemical inertness were a priority, dry nitrogen, or at worst, argon (only needed for the most reactive metals, like welding titanium for example), would do the job.

Indeed, argon is used for this reason in incandescent lamps: reduced convection loss, compared to other gasses.  Krypton is even better (heavier, slower), and xenon still more; but they are relatively expensive, so don't see widespread use, except where very small quantities are needed ("krypton" flashlight bulbs), or where the properties are directly required (xenon flash and HID lamps are excellent because xenon has a broad "white" response with few bright spectral lines; obviously, the properties of ionized xenon being somewhat more important than its gaseous properties!).

Tim

[kc8apf]

HGST talks about this in their helium technology brochure (http://www.hgst.com/sites/default/files/resources/HGST-Helium-Technology-BR03.pdf).  Helium, being less dense than air, reduces turbulence and allows the platters to spin more easily at the same speeds.  Someone asked about what happens if a drive leaks.  There is no built-in sensor for pressure change or chemical detection.  The drive just slows down and performs poorly if enough helium is lost.

[rs20]

HGST talks about this in their helium technology brochure (http://www.hgst.com/sites/default/files/resources/HGST-Helium-Technology-BR03.pdf).  Helium, being less dense than air, reduces turbulence and allows the platters to spin more easily at the same speeds.  Someone asked about what happens if a drive leaks.  There is no built-in sensor for pressure change or chemical detection.  The drive just slows down and performs poorly if enough helium is lost.

Surely the spindle is driven by a motor with feedback, would it not just push harder to attain the engineered speed? I'd be surprised if the change caused by a 10% infiltration of nitrogen would be significant compared to the lubrication being cold, or a tiny bit of grit in the bearings, etc. They wouldn't cut the design that fine.

They talk about allowing the heads to fly closer to the disk due to reduced turbulence, based on this my conclusion would be that there would be an increased risk of a head crash if a) the drive leaked and b) the driver had no way of detecting the leak and explicitly slowing down in response. Do you have a source to justify your claim that the platter will spin more slowly if there's a leak? Or are you referring to the arm swinging around more slowly?

[matseng]

I'd be surprised if the change caused by a 10% infiltration of nitrogen would be significant compared to the lubrication being cold, or a tiny bit of grit in the bearings, etc. They wouldn't cut the design that fine.

If the enclosure is filled 100% pure helium and it is solid enough (aluminium?) to not let "regular" gasses diffuse through it but permeable enough to have the helium escape, what would then happen?  Would the pressure inside just drop over time until all helium is gone and there's just a vacuum there? Or would an equilibrium point be reached after a while?

[grumpydoc]

I'd be surprised if the change caused by a 10% infiltration of nitrogen would be significant compared to the lubrication being cold, or a tiny bit of grit in the bearings, etc. They wouldn't cut the design that fine.

If the enclosure is filled 100% pure helium and it is solid enough (aluminium?) to not let "regular" gasses diffuse through it but permeable enough to have the helium escape, what would then happen?  Would the pressure inside just drop over time until all helium is gone and there's just a vacuum there? Or would an equilibrium point be reached after a while?

Presumably it would leak until the vapour pressure of helium inside is equal to that outside.

[rs20]

Presumably it would leak until the vapour pressure of helium inside is equal to that outside.

Which would correspond to an almost perfect vacuum -- I'm skeptical, I feel like I can maybe build a perpetual motion machine if this were true.

[Rick Law]

Presumably it would leak until the vapour pressure of helium inside is equal to that outside.

Which would correspond to an almost perfect vacuum -- I'm skeptical, I feel like I can maybe build a perpetual motion machine if this were true.

Not possible for it to leak to almost perfect vacuum.  Once the pressure inside is less than the outside world, air will leak in to fill the void.

Unless of course if the drive is operating in space in which case it could indeed leaks down to near vacuum...

[matseng]

Not possible for it to leak to almost perfect vacuum.  Once the pressure inside is less than the outside world, air will leak in to fill the void.

That doesn't sound right.  You're saying that it's impossible to keep a vacuum (or near vacuum) inside a sealed canister?  Lucky for us born before the advent of flat-TVs that the CRT tubes didn't know about that.  :-D

[rs20]

Not possible for it to leak to almost perfect vacuum.  Once the pressure inside is less than the outside world, air will leak in to fill the void.

Unless of course if the drive is operating in space in which case it could indeed leaks down to near vacuum...

Right, but we're talking about a (hypothetical?) material here which allows Helium to diffuse through, but not N2 or O2. It seems weird to claim that the helium would leak out and leave a vacuum behind, I agree. But it also seems weird to assert that the high pressure of N2 and O2 on the outside would somehow prevent He from diffusing out, even though the N2 or O2 can't diffuse into the material in order to provide this preventative force?

What are you claiming will happen to a vessel constructed of a material permeable to Helium, but impermeable to everything else?

[kc8apf]

HGST talks about this in their helium technology brochure (http://www.hgst.com/sites/default/files/resources/HGST-Helium-Technology-BR03.pdf).  Helium, being less dense than air, reduces turbulence and allows the platters to spin more easily at the same speeds.  Someone asked about what happens if a drive leaks.  There is no built-in sensor for pressure change or chemical detection.  The drive just slows down and performs poorly if enough helium is lost.

Surely the spindle is driven by a motor with feedback, would it not just push harder to attain the engineered speed? I'd be surprised if the change caused by a 10% infiltration of nitrogen would be significant compared to the lubrication being cold, or a tiny bit of grit in the bearings, etc. They wouldn't cut the design that fine.

They talk about allowing the heads to fly closer to the disk due to reduced turbulence, based on this my conclusion would be that there would be an increased risk of a head crash if a) the drive leaked and b) the driver had no way of detecting the leak and explicitly slowing down in response. Do you have a source to justify your claim that the platter will spin more slowly if there's a leak? Or are you referring to the arm swinging around more slowly?

I wasn't clear.  By using helium, the energy required to spin the platter at a constant speed is lower.  If the helium leaks (and is replaced by air), the motor controller will compensate to keep a constant speed but will use more energy.

The other major effect of a helium leak is that the increased turbulence causes the drive to fly the heads higher to avoid impacts.  This causes a higher rate of read errors which means longer read command times (drives internally retry up to 10+ times before returning failure to the host).  The net effect is that data throughput decreases once enough helium has leaked out.

I can't point you to a source as this was discussed in person with the product manager and lead engineer of the helium project.

[Rick Law]

Not possible for it to leak to almost perfect vacuum.  Once the pressure inside is less than the outside world, air will leak in to fill the void.

That doesn't sound right.  You're saying that it's impossible to keep a vacuum (or near vacuum) inside a sealed canister?  Lucky for us born before the advent of flat-TVs that the CRT tubes didn't know about that.  :-D

You misunderstood what I wrote.  To better express what I mean, I have to use all the words however long the sentence: to leak air from inside that sealed space to the outside, leaking enough to a point that the inside is almost perfect vacuum.  That is impossible.  The inside cannot have lower air pressure than the outside and still be leaking air outward.

[rs20]

The other major effect of a helium leak is that the increased turbulence causes the drive to fly the heads higher to avoid impacts.

I may be nitpicking on your precise choice of words here, but the drive doesn't fly the heads, nor does it have any control over the height of the heads, right? The head spacing is maintained entirely through the wing-on-ground effect, at least that was my understanding.

If you're saying that the increased density causes the head to simply be pushed further away, then yeah everything you say makes sense.

[rs20]

The inside cannot have lower air pressure than the outside and still be leaking air outward.

There's still something weird going on here, as I expressed in my earlier question to you. Consider a 0.5 atmosphere Helium atmosphere, in a Helium-only-permeable vessel, sitting in normal 1 atmosphere air. So we've got a Helium atom, that's diffusing happily through the bulk of the vessel wall. It reaches the outside interface of the vessel wall, where it meets atmospheric air. Does the Helium get turned back; prevented from actually completely leaking out of the vessel by some unknown force that is omnipotently aware that the vessel from which it came is at a lower pressure than the air which it is about to enter? Or does it escape, leading to a situation which asymptotically leads to the vessel becoming more and more depressurised? I totally get that both possibilities I present here seem absurd, but I can't tell what kind of third story I can tell about any given Helium atom that makes any sense.

[grumpydoc]

As far as I understand it is the partial pressure of the individual gasses which matter. So, if the vessle is permiable to Helium, it will leak out until the pressure of helium inside is equal to the partial pressure of helium outside (not very much for atmostpheric helium).

Of course O₂ and N₂ mollecules are trying to leak in at the same time.