Eagerly waiting for review, I would like to know how fast that helium is going to diffuse through those seals. Helium is quite notorious when it comes to diffusion through other materials.
I was wondering about the same thing. But I guess that if He was to leak out, that would only make the drive perform better, since gas density and drag would be reduced even more. Presumably the seals are tight enough to prevent N2 and O2 from leaking in, which is far easier to achieve than preventing He to leak out.
Well if He does leak, it would be replaced by the outside air. The air density would remain about the same. The drive would slowly die out, but I think up to that point the drive would be dead by some other unrelated causes anyway.
They've been working on that seals for about 10 years before introducing the 1st drives in 2013. It's a fair guess to assume they're OK by now. At least I haven't heard of any leakage problems with the initial HGST drives.
Thanks for the part of your comment that was constructive : ]
I found this article on Forbes that explains why there isn't really a shortage, good to know! Didn't find this article in the quick search I did before my original post.
1) the helium is in there under pressure. It might be, but not much pressure probably. Think about it, if it were under pressure, excess would probably leak out until it's not under pressure. But, it would still be 100% helium in there. In fact, they probably don't want it under pressure as that would increase the density and it would lose some of it's advantages over air.
2) Just because helium might even be able to leak out, it doesn't mean that air can leak IN. First, the atoms are significantly bigger. Second, with the seal in there, it not like it's exchanging and mixing air for helium freely...
I think you meant that He (Helium) atoms are smaller than the molecules in the air. They also won't chemically react with anything the drive is built with.
Well, he meant the atoms in the air are larger than the ones in He, so the ambient air is unlikely to leak INto the drive.
Although I would assume they would use He at a similar pressure to ambient air, so there would be no reason for the He to leak out or regular air to leak in.
You don't understand the physics. Just like osmosis, the helium atoms diffuse to where there are less helium atoms. So they will diffuse out even if equal pressure until the loss in pressure equals the "pressure" to drive them out. As was stated in another post, the air will not diffuse/leak into the drive because Nitrogen and the other components of air do not diffuse through materials as easily. Now if there was an actual defect casuing a leak, then the air would get in and the helium escape.
Maybe they can expect it because both WD/HGST and Seagate spent nearly a decade developing the HelioSeal technology, and millions of the drives have been shipped.
Considering the massive challenges of semiconductor feature size reduction at the moment, the Hard Drive manufacturers have nothing to fear at the very least until EUV lithography comes online commercially in 3-8 years.
NAND probably will never catch up to hard drives in price per capacity due to the massive loss in durability when lowering lithographic nodes as well.
The gap will most likely be filled by the NAND successor, RRAM (3D XPoint)
When comparing against the lowest tier drives, maybe. But WD Reds, Blacks, Purples and Enterprise drives are pretty pricey compared to these and I don't see NAND remaining much higher than them for long. From what I remember the original 850 Evos were on a 40nm node and SSDs are down to what, 32c USD/GB?
That's quite a lot of room for die/cost shrinks, and Samsung has already proven excess density isn't the issue with that 16TB concept in 2.5". They're not scraping margins, and if they can increase volume at a lower cost/GB by adding more density they could lower their margins but increase overall profitability without needing to ship any more units.
At that point they'd likely find they do ship a lot more units. Enterprise is looking pretty closely at them as well, the sheer IOPS from them is just completely incomparable to a HDD, and I'm sure Samsung is carefully weighing its current margins to a theoretical "close-enough" point when it would be silly to buy an HDD. I wouldn't be surprised if they suddenly went from 20c/GB to 10c/GB in one big jump when they finally hit 20.
What you don't understand is that 15nm/19nm Toshiba is the lowest cost node until Micron and Toshiba up their game on 3D NAND stacking to put pricing pressure on Samsung.
3D NAND stacking also doesn't lower prices in the same way that node shrinks do, as the wafers per day is not equivalent, therefore the areal 2D density advantage of 3D NAND doesn't factor in the same way.
TL/DR: Moore's law died for realzies this time, in everything.
I was suspicious as most in IT are with new technology when HGST released the Ultrastar HE, but we've had pretty good luck with them at various clients in NAS applications. We had one fail in a RAID5 array on a mail server with 6 drives, but had a spare on-site so swapping in was no problem, and this isn't terribly unusual. I wasn't there to see it, and it was shipped in for a warranty exchange before I got to plug it in and see what SMART errors there were and run any tests. Basically the RAID array said it failed and the guy on-site pulled it immediately.
To this day I'm curiously exactly what went wrong with it, if anything was wrong with it at all. Sometimes these Areca RAID cards just kick a disk out that's fine and just needs an integrity check, especially under sustained 24/7 workloads like a mail server (that was also running a SQL database, so basically high demand.) This was April 2015 and they were put into service in March 2014.
I'm excited to see these hit the consumer space. I'd rather have a hermetically sealed drive than a shingled recording drive...time will tell exactly how reliable SMR is, just like time has proven the He6 is.
I have historically dealt with 30 gb to 2 T databases, adn doing all-flash for that is prohibitively expensive, expecially if you are mirroring the whole thing for perfect redundancy (a necessity with my crap).
The only time I see all-flash is where yuo need speed, and even then they have those huge RAM caches. Personally i have seen large RAID arrays with gib ole spinning drives do amazingly well withe those huge RAM sans. They are faster than all flash. Those mixture HDD, SSD, and RAM multi-level cache thingy's are the only ones I have never seen in action. They do look like they would solve a lot, but for lots of stuff simply having insanely fast RAM caches are more than adequate for big RAID storage for DB's. At least for me I tend to see usage where there are heavily-used small DB's or heavily used views or reports made off of views and the entire high-usage areas are only 30-40gb. If you have > 60gb RAM cache, the rest if kind of irrelevant. The idea of having a terabyte flash layer in between seems like a waste.
while i agree that SSDs are obviously not always the best choice for a database cluster (reasons are fairly obvious -- cost and capacity mainly), your examples aren't really helpful.
cache is only good for reads.. or if you aren't flushing to disk on transactions (in other words you dont care about your data). even then, only if your data set fits in memory.
for a performant db backed by magnetic storage just make sure to use plenty of spindles.
Because it's a waste of money on a mail server. The I/O's are extremely low.
The real question is why RAID5 as it's the worst possible configuration. The chance for URE after 3 years is a very bad percentage. The rebuild time on 8TB hard drive is extremely long, not to mention the resliver operation can degrade performance drastically.
Like I said before, from what I remember the node for each layer was 40ish on the 850, there was a teardown somewhere, maybe here? If they can do node shrinks now, as far down as 15nm then they needn't add more layers for increased density and cost reduction.
Also my replies appear to keep creating new threads on the reply button, IE. Hopefully this pops as a reply.
The widespread use of He is of some concern since the stuff is difficult to come by and national reserves are not doing well. I wish there were a practical way to synthesize the stuff.
Hlpeful typo reprot: "WD is not disclosing much [int he] way of technical details about its new hard drives"
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
34 Comments
Back to Article
PrinceGaz - Tuesday, March 1, 2016 - link
Helium-filled. What happens if you hold one at arms-length and let go? :)Ryan Smith - Tuesday, March 1, 2016 - link
It comes crashing down and you've broken a $300 hard drive.UltraWide - Tuesday, March 1, 2016 - link
LOL oops!wilhil - Tuesday, March 1, 2016 - link
Your data goes to the cloud!osxandwindows - Tuesday, March 1, 2016 - link
saw what you did there.:d
Chaitanya - Tuesday, March 1, 2016 - link
Eagerly waiting for review, I would like to know how fast that helium is going to diffuse through those seals. Helium is quite notorious when it comes to diffusion through other materials.nathanddrews - Tuesday, March 1, 2016 - link
Not unlike argon-filled windows. My guess is that the time it takes to leak out will exceed the life of the drive. One can hope, anyway.JanW1 - Tuesday, March 1, 2016 - link
I was wondering about the same thing. But I guess that if He was to leak out, that would only make the drive perform better, since gas density and drag would be reduced even more. Presumably the seals are tight enough to prevent N2 and O2 from leaking in, which is far easier to achieve than preventing He to leak out.mr_tawan - Tuesday, March 1, 2016 - link
Well if He does leak, it would be replaced by the outside air. The air density would remain about the same. The drive would slowly die out, but I think up to that point the drive would be dead by some other unrelated causes anyway.DanNeely - Tuesday, March 1, 2016 - link
If the ambient air doesn't leak in as fast as He leaks out the pressure difference will crush the drive.MrSpadge - Tuesday, March 1, 2016 - link
They've been working on that seals for about 10 years before introducing the 1st drives in 2013. It's a fair guess to assume they're OK by now. At least I haven't heard of any leakage problems with the initial HGST drives.valinor89 - Tuesday, March 1, 2016 - link
Has someone even investigated the long term composition of the gas inside those hard drives? Might be a challenging but interesting project.Rocket321 - Tuesday, March 1, 2016 - link
My helium drive makes all my mp3's sound like Alvin and the Chipmunks.ayejay_nz - Tuesday, March 1, 2016 - link
I thought the world was running out of helium?If so, this seems like a waste of the resource!
nandnandnand - Tuesday, March 1, 2016 - link
Well you were wrong. The helium shortage is economic, and these use barely any helium.Stop repeating bad information.
ayejay_nz - Tuesday, March 1, 2016 - link
Thanks for the part of your comment that was constructive : ]I found this article on Forbes that explains why there isn't really a shortage, good to know! Didn't find this article in the quick search I did before my original post.
http://www.forbes.com/sites/timworstall/2015/06/18...
Shadowmaster625 - Tuesday, March 1, 2016 - link
I dont see how they can expect the helium to stay in there for more than a few months.bill.rookard - Tuesday, March 1, 2016 - link
Well, you're assuming that:1) the helium is in there under pressure. It might be, but not much pressure probably. Think about it, if it were under pressure, excess would probably leak out until it's not under pressure. But, it would still be 100% helium in there. In fact, they probably don't want it under pressure as that would increase the density and it would lose some of it's advantages over air.
2) Just because helium might even be able to leak out, it doesn't mean that air can leak IN. First, the atoms are significantly bigger. Second, with the seal in there, it not like it's exchanging and mixing air for helium freely...
Fujikoma - Tuesday, March 1, 2016 - link
I think you meant that He (Helium) atoms are smaller than the molecules in the air. They also won't chemically react with anything the drive is built with.extide - Tuesday, March 1, 2016 - link
Well, he meant the atoms in the air are larger than the ones in He, so the ambient air is unlikely to leak INto the drive.Although I would assume they would use He at a similar pressure to ambient air, so there would be no reason for the He to leak out or regular air to leak in.
megalee - Tuesday, March 1, 2016 - link
You don't understand the physics. Just like osmosis, the helium atoms diffuse to where there are less helium atoms. So they will diffuse out even if equal pressure until the loss in pressure equals the "pressure" to drive them out. As was stated in another post, the air will not diffuse/leak into the drive because Nitrogen and the other components of air do not diffuse through materials as easily. Now if there was an actual defect casuing a leak, then the air would get in and the helium escape.nandnandnand - Tuesday, March 1, 2016 - link
Maybe they can expect it because both WD/HGST and Seagate spent nearly a decade developing the HelioSeal technology, and millions of the drives have been shipped.linker3000 - Wednesday, March 2, 2016 - link
I'll just go ask the 6 x 6TB He6 drives in one of my servers - that's the disks date stamped May and June 2013.Danvelopment - Tuesday, March 1, 2016 - link
Does that mean price/GB on regular drives are dropping yet? Or do they still remain unchanged from 2012?Their unwillingness to lower prices in the face of such rapidly dropping NAND prices is what will eventually seal their fate.
In the same time NAND has gone from 40x more costly per GB to 7x more costly today.
Communism - Tuesday, March 1, 2016 - link
Considering the massive challenges of semiconductor feature size reduction at the moment, the Hard Drive manufacturers have nothing to fear at the very least until EUV lithography comes online commercially in 3-8 years.NAND probably will never catch up to hard drives in price per capacity due to the massive loss in durability when lowering lithographic nodes as well.
The gap will most likely be filled by the NAND successor, RRAM (3D XPoint)
Danvelopment - Wednesday, March 2, 2016 - link
When comparing against the lowest tier drives, maybe. But WD Reds, Blacks, Purples and Enterprise drives are pretty pricey compared to these and I don't see NAND remaining much higher than them for long. From what I remember the original 850 Evos were on a 40nm node and SSDs are down to what, 32c USD/GB?That's quite a lot of room for die/cost shrinks, and Samsung has already proven excess density isn't the issue with that 16TB concept in 2.5". They're not scraping margins, and if they can increase volume at a lower cost/GB by adding more density they could lower their margins but increase overall profitability without needing to ship any more units.
At that point they'd likely find they do ship a lot more units. Enterprise is looking pretty closely at them as well, the sheer IOPS from them is just completely incomparable to a HDD, and I'm sure Samsung is carefully weighing its current margins to a theoretical "close-enough" point when it would be silly to buy an HDD. I wouldn't be surprised if they suddenly went from 20c/GB to 10c/GB in one big jump when they finally hit 20.
Communism - Wednesday, March 2, 2016 - link
What you don't understand is that 15nm/19nm Toshiba is the lowest cost node until Micron and Toshiba up their game on 3D NAND stacking to put pricing pressure on Samsung.3D NAND stacking also doesn't lower prices in the same way that node shrinks do, as the wafers per day is not equivalent, therefore the areal 2D density advantage of 3D NAND doesn't factor in the same way.
TL/DR: Moore's law died for realzies this time, in everything.
Samus - Wednesday, March 2, 2016 - link
I was suspicious as most in IT are with new technology when HGST released the Ultrastar HE, but we've had pretty good luck with them at various clients in NAS applications. We had one fail in a RAID5 array on a mail server with 6 drives, but had a spare on-site so swapping in was no problem, and this isn't terribly unusual. I wasn't there to see it, and it was shipped in for a warranty exchange before I got to plug it in and see what SMART errors there were and run any tests. Basically the RAID array said it failed and the guy on-site pulled it immediately.To this day I'm curiously exactly what went wrong with it, if anything was wrong with it at all. Sometimes these Areca RAID cards just kick a disk out that's fine and just needs an integrity check, especially under sustained 24/7 workloads like a mail server (that was also running a SQL database, so basically high demand.) This was April 2015 and they were put into service in March 2014.
I'm excited to see these hit the consumer space. I'd rather have a hermetically sealed drive than a shingled recording drive...time will tell exactly how reliable SMR is, just like time has proven the He6 is.
Communism - Wednesday, March 2, 2016 - link
What's the point of even running databases on 8TB hard drives in RAID5 instead of just going the SSD route?Either you are stacking ungodly amounts of DRAM as cache or you are just doing extremely large basically never accessed datasets?
SkipPerk - Monday, March 7, 2016 - link
I have historically dealt with 30 gb to 2 T databases, adn doing all-flash for that is prohibitively expensive, expecially if you are mirroring the whole thing for perfect redundancy (a necessity with my crap).The only time I see all-flash is where yuo need speed, and even then they have those huge RAM caches. Personally i have seen large RAID arrays with gib ole spinning drives do amazingly well withe those huge RAM sans. They are faster than all flash. Those mixture HDD, SSD, and RAM multi-level cache thingy's are the only ones I have never seen in action. They do look like they would solve a lot, but for lots of stuff simply having insanely fast RAM caches are more than adequate for big RAID storage for DB's. At least for me I tend to see usage where there are heavily-used small DB's or heavily used views or reports made off of views and the entire high-usage areas are only 30-40gb. If you have > 60gb RAM cache, the rest if kind of irrelevant. The idea of having a terabyte flash layer in between seems like a waste.
skrewler2 - Thursday, March 10, 2016 - link
while i agree that SSDs are obviously not always the best choice for a database cluster (reasons are fairly obvious -- cost and capacity mainly), your examples aren't really helpful.cache is only good for reads.. or if you aren't flushing to disk on transactions (in other words you dont care about your data). even then, only if your data set fits in memory.
for a performant db backed by magnetic storage just make sure to use plenty of spindles.
Dug - Monday, March 7, 2016 - link
Because it's a waste of money on a mail server. The I/O's are extremely low.The real question is why RAID5 as it's the worst possible configuration.
The chance for URE after 3 years is a very bad percentage. The rebuild time on 8TB hard drive is extremely long, not to mention the resliver operation can degrade performance drastically.
Danvelopment - Thursday, March 3, 2016 - link
How about 3D stacking and node shrinks?Like I said before, from what I remember the node for each layer was 40ish on the 850, there was a teardown somewhere, maybe here? If they can do node shrinks now, as far down as 15nm then they needn't add more layers for increased density and cost reduction.
Also my replies appear to keep creating new threads on the reply button, IE. Hopefully this pops as a reply.
Sivar - Monday, March 7, 2016 - link
The widespread use of He is of some concern since the stuff is difficult to come by and national reserves are not doing well. I wish there were a practical way to synthesize the stuff.Hlpeful typo reprot:
"WD is not disclosing much [int he] way of technical details about its new hard drives"