SSD Latency, Error Rates May Spell Bleak Future 292
Lucas123 writes "A new study by the University of California and Microsoft shows that NAND flash memory experiences significant performance degradation as die sizes shrink in size. Over the next dozen years latency will double as the circuitry size shrinks from 25 nanometers today, to 6.5nm, the research showed. Speaking at the Usenix Conference on File and Storage Technologies in San Jose this week, Laura Grupp, a graduate student at the University of California, said tests of 45 different types of NAND flash chips from six vendors using 72nm to 25nm lithography techniques showed performance degraded across the board and error rates increased as die sizes shrunk. Triple-Level NAND performed the worst, followed by Multi-Level Cell NAND and Single-Level Cell. The researchers said MLC NAND-based SSDs won't be able to go beyond 4TB and TLC-based SSDs won't be able to scale past 16TB because of the performance degradation, so it appears the end of the road for SSDs will be 2024."
Sounds legit (Score:5, Insightful)
Re:Sounds legit (Score:5, Funny)
OK then. You've got 10 years. Get going.
Re:Sounds legit (Score:5, Funny)
Re:Sounds legit (Score:5, Funny)
Would those be chipmonks?
OK, OK, I'm going........
Re:Sounds legit (Score:5, Insightful)
Re: (Score:3)
Why? I'd rather have a bunch of 80gig disks in a RAID 50 it would be much faster than a 2TB drive and far more stable in case of data loss.
I guess if what matters to you is small size and your data has little value, a single 2TB drive is good. to me the data is worth far more than the price of the hardware.
Re:Sounds legit (Score:5, Insightful)
Because 25 hard drives would be a bitch to carry around in your laptop?
Re: (Score:3)
Your sig reminds me of my reply, "I went with the hard drives. I find the seek times on Netflix unacceptable."
Re: (Score:3)
Re:Sounds legit (Score:5, Informative)
Well, to start with you can make an SSD as big as you want by taking smaller SSD's and chaining them together with an intelligent front-end.
I could do the same thing with a bunch of 80 GB hard disks, but I'd rather just buy a 2 TB one and run that instead.
Did you know that your hard disk is actually already made out of multiple platters with smaller capacities that make up the whole transparently? Your RAM is made up of dozens of individual smaller chips that make up the total capacity, and so are existing SSDs and USB flash memory sticks.
Kids these days.
Re:Sounds legit (Score:5, Insightful)
Oh there will be a great discovery/invention in the next 10 years. Unfortunately it will be tied up in patent litigation for the next 50 years after that. All fun and games when it is a hard drive. Not so funny when it is a medicine that can save your kid.
Re:Sounds legit (Score:5, Funny)
Re: (Score:3, Insightful)
HDD tech has advanced without patent litigation tying anything up. What makes you think it will be different for NAND's successor?
Re: (Score:3)
Re:Sounds legit (Score:5, Interesting)
We already have the breakthrough, but it's not Flash, it's PRAM [wikipedia.org].
Re:Sounds legit (Score:5, Funny)
Yes, but what I heard about PRAM is that you have to push it. A lot.
Re:Sounds legit (Score:5, Funny)
But I *LIKE* to push the PRAM a lot!
Re:Sounds legit (Score:5, Funny)
On behalf of my fellow Brits I would like to apologise, and assure you that henceforth we shan't abbreviate the full term, perambulator. Let's face it: pushing an overloaded baby carriage (including baby, nappy bags, bottles, snacks, toys, etc.) is a long way from being the "leisurely walk" for which the word "stroller" would be appropriate.
Re: (Score:3)
That's fine, just so long as the editor who allowed it to be abbreviated is sacked.
Re:Sounds legit (Score:4, Insightful)
It would help if those damn Brits wouldn't use such ridiculous words. Here on the reasonable side of the Atlantic, we call them strollers.
Understand I'm a North American. So when I point out your wrongness, it's not because I'm a damned Brit.
Sprinters sprint.
Runners run.
Juggers jog.
You'd think strollers stroll.
Strangely instead strollers convey small children nestled within their confines but only because someone is pushing them along.
Know what. I'll take a word that is allocated to naming the object any day over a misleading one. The Brits got it right, we got it wrong.
Speaking of wrong, the word you're looking for is "buggy". Like... rubber baby buggy bumpers.
Re:Sounds legit (Score:5, Funny)
Juggers jog.
Joggers jog. Juggers bounce.
Re:Sounds legit (Score:4, Informative)
Re:Sounds legit (Score:5, Informative)
We already have the breakthrough, but it's not Flash, it's PRAM [wikipedia.org].
And MRAM. [wikipedia.org] And FeRAM. [wikipedia.org]
Re: (Score:3)
Samsung managed 1 Gbit at 58 nm in February 2011. The rest of the alternatives are significantly lower density than even PRAM. Not particularly promising IMO.
Re: (Score:3)
That's easy, you just ignore the past 12 years or so, and just look at flight from its beginning to c.2000. It's not like we know what computing will look like in 50 years; for all we know, that could also turn into a disaster, with a free and open internet being a thing of the past, and government firewalls keeping track of everything you do online. Everything has the potential to be totally fucked up or ruined altogether. Just look at hitchhiking. It used to be an acceptable way of getting around at o
Re:Sounds legit (Score:5, Interesting)
Perhaps it's already been found:
http://en.wikipedia.org/wiki/Phase-change_memory [wikipedia.org]
PCM still has hurdles to overcome, but it's generally considered that performance increases as size decreases, the opposite of NAND.
Re: (Score:2)
Because there could *never* be a breakthrough discovery/invention found within the next 10 years.
Tomorrow's "breakthrough" doesn't mean you have a commercially viable product within the next ten years --- or even the next twenty,
Re:Sounds legit (Score:5, Insightful)
what about the last 3-4 years worth of discoveries, of phase change, memresistors, etc. Many of which get more efficient the smaller you go.
So NAND Flash has a lifespan. big deal, So did magnetic core drives, Hard drives are still going strong but are reaching the top ends for themselves too.
Re:Sounds legit (Score:5, Interesting)
They all have much lower densities. The highest is PRAM at 1 Gbit with a 58 nm process, demonstrated by Samsung in February 2011. That's way too low.
Re:Sounds legit (Score:5, Insightful)
Re: (Score:3, Insightful)
COBOL
Re: (Score:3)
Wasn't there about 6 alternatives to NAND discovered last year? I think IBM announced 2 of them.
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Memresistors are suppose to come out in the next 1-2 years. Will be even used as system memory because it has no effective "wear". That "breakthrough" is already done, it's just being readied for production.
Re:Sounds legit (Score:4, Interesting)
Didn't you hear? We've reached the limitations of technology and innovation.
That's why it's so stupid to put any money into non-fossil energy. If we can't power a house by solar energy now, we'll never be able to and we just have to accept it.
It's the End of History. Again.
Re: (Score:3)
No, no, I'm being a smart ass. We've altered out house with both passive and active solar systems, too. It's not 100% but we pay very small electric and heating bills compared to our neighbors, and we live in the wintry midwest (Chicago).
I believe that it's stupid to put any more money into fossil fuel technologies until we've got some serious dedication to developing alternatives, not just lip service a
Re: (Score:3)
Re: (Score:3)
And to make matters worse they don't "fail gracefully" as the old spinning rust does. honestly i can't remember a HDD that failed without warning in the past....oh hell the last one was probably a Deathstar around 2000, no thanks to SMART you'll usually get SOME kind of indication, be it SMART or noise or weird errors, something, and then you can get your data off. My gamer customers went back to running raptors in RAID because they bought SSDs and lost data, just one day they flipped the switch and poof! No drive even in BIOS and no way for me to get a single byte of data back.
FLASH does (should) indeed fail gracefully. Once a block wears out, programming it will fail, and the FLASH and controller will know this and mark it bad. But other blocks will still be readable, and the now dead block contained no useful data (else the controller wouldn't be erasing it.)
What you're talking about are firmware based bugs, the controller not making the FLASH contents accessible. These problems are probably the result of block translation tables being corrupted, and is entirely the fault of th
SSD =/= NAND Flash (Score:5, Informative)
Re: (Score:3)
The advantage of cheapness is you can use more to buy reliability. It's what Google does with servers, for example. And, for that matter, SSDs already do it with ECC and CRC. As the reliability goes down, you just add more. And add some extra capacity too - that helps with both wear leveling and provides spare blocks for when they start failing. And for storage, it's really easy to swap in a spare. In many applications (RAM, cache, etc.), there are a few extra rows of storage to tolerate manufacturing defec
In other news... (Score:4, Informative)
Re: (Score:2)
....conflating what's possible with what's desired.
One of these depends on human nature and the other one depends on physics.
nature always finds a way (Score:2)
Re:In other news... (Score:5, Insightful)
"Hey, would you want a computer? It can fit in your pocket, let you talk to anyone in the world, can take pictures and provide you god damn near any information written down by a human being, and you can watch porn on it!"
Computers are the same thing they were even 20 years ago in name only.
Re: (Score:3)
You're not looking at it from the right level of abstraction.
The information of yesterday was Enigma codes. The information of today is pictures, wikipedia, and pr0n. The computers of yesterday and today are effectively doing the same things: storing and moving information and making calculations to glean new information.
They are the same. You just can do more with it now, because your potential is limited, while a computer's potential is only limited by technological progress. But the theory backing comput
Re: (Score:3, Funny)
A year? Yeah, I could live pretty well off of that...
Re: (Score:3)
Actually, there is no credible source on that...
Re: (Score:3)
FWIW, it wasn't a study, it was an off the cuff statement by some major wheel at IBM. (I'm not sure it was Watson.) At the time it was a true statement. It didn't stay true.
There *is* actual historical evidence, even though I don't have it. But googling for '"only 6 computers" IBM' yielded among other links: ... Aiken thought that only 6 comput
History of Computing timeline | Timetoast timelines
www.timetoast.com/timelines/33262
03/01/1939, IBM sponsors an engineer, Howard Aiken, tried to intergrate 73 IBM
Stuff like this... (Score:5, Insightful)
... always denies other areas of innovation. The same way processors were thought not to scale down to x nm and we're at 20'ish nm now. The same way hard drives were thought only to have x capacity and we're now in the terabytes. If nand is really so limited then something different then nand will take it's place. But a few terabyte will be more then enough for 99% of applications and hard disks will be for packrats and those who need large amounts of longer term storage.
Re: (Score:2)
One time, my wife's cousin (who was studying RAM at MIT and is now a brain specialist teaching at Stanford) said that "you will never be able to put more than 40MB on a PCMCIA card."
I replied, "Within 5 years, we'll be carrying a GB around in our pocket the size of a postage stamp." I was right. Sometimes smart guys are so focused on their area that they fail to see the realities of supply and demand combined with Moore's law.
Re:Stuff like this... (Score:5, Insightful)
"640k ought to be enough for everyone!"
One can take a look at videocards, right now for most PC gamers they haven't needed to upgrade their video hardware for quite some time relatively speaking compared to the past. The idea that needs will scale linearly forever is nonsense.
There is a point after rapid growth where you reach 'good enough' until the next step is ready which no one knows in advance.
Re: (Score:2)
That's funny because I just tried to install a Linux RTS game on a couple of my other machines. Neither one of them was equipped to deal with it. "Better drivers" on Windows didn't help the Win7 box. I didn't even try the Mac; it was trailing edge when it was new.
Good enough changes year by year and people find new ways to exploit hardware that some people might have thought is "tapped out".
VDPAU and VAAPI are both good examples of that.
I would still ditch pretty much any embedded GPU on the market today an
Re:Stuff like this... (Score:4, Insightful)
Obligatory ... (Score:2, Redundant)
Re: (Score:2)
"16TB ought to be enough for anybody."
Didn't you mean "640TB blah blah blah".
We've already got at least 20TB of fixed disks at home (including online backups). The media server alone has 12TB.
I want HAL's memory (Score:4, Interesting)
Re: (Score:2, Funny)
Holographic memory requires fusion power.
Re: (Score:3)
Not a problem, since I have that in my goddamned flying car.
Re:I want HAL's memory (Score:5, Funny)
A warp core really isn't a power source. It is more like an alternator. The power source is the matter-antimatter reactions. Similarly people confuse dilithium crystals with being a power source when they are really just a matter-antimatter regulator.
And now, back to reality...
Re: (Score:3)
Re:I want HAL's memory (Score:5, Funny)
So... How's your virginity going?
Re:I want HAL's memory (Score:4, Funny)
Still waiting for the Holographic Memory that should have been hear a decade ago.
- there is the problem.
With holographic memory you shouldn't be trying to 'hear' anything, it's something likely in visible electromagnetic spectrum instead!
Re: (Score:2)
Holographic memory is capacious enough, but it's SLOW. And nobody's come up with a way to make it cheaply. (Much less to make it Read/Write & cheaply.)
Re: (Score:2)
I worked on holographic memory. It has a huge capacity, but very very slow write times. It was something like 1 byte per second, or something ridiculous like that.
If people could come up with a medium that could be developed quickly, it could be neat.
will vertical 3D NAND flash be here in 2013 (Score:2)
"...the end of the road for SSDs will be 2024..." (Score:5, Insightful)
Yes. They'll all stop working then and it will become impossible to make any more.
Re: (Score:3)
Well, not so much that but rather than hard drive rotational latencies will finally catch up to nand. With our disks spinning at a paltry 100,000,000 rpm, latency will finally be a worry of the past.
Re: (Score:2)
I've wondered what spinning disk could do in a vacuum chamber and with a non-contact magnetic bearing.
Re: (Score:2)
You still have the limits of the disk to deal with. That's why optical media like DVDs and CDs aren't getting any faster. The disks are already spinning as fast as they can.
Not bleak at all (Score:2, Informative)
From the article, "This will reduce the write latency advantage that SSDs offer relative to disk from 8.3x (vs. a 7 ms disk access) to just 3.2x.". Yeah, doom and gloom.
Re:Not bleak at all (Score:5, Informative)
Throughput isn't measured in IOPS (Score:2)
Actually, that's when I realized that the guy writing the article didn't have a clue. Since when is throughput measured in IOPS?
Re: (Score:3)
Actually, that's when I realized that the guy writing the article didn't have a clue. Since when is throughput measured in IOPS?
Since always. Throughput is always operations per second, or transactions per second. Bandiwdth is measred in Mbps or MBps.
I'm sure it's all wonderful (Score:5, Funny)
http://cseweb.ucsd.edu/~lgrupp/CV.pdf [ucsd.edu]
4TB limit (Score:5, Insightful)
Re: (Score:2)
I hate to break it to you but that is 8 drives in one device. Hence the "octal" name.
Re: (Score:2)
They're talking about on a single device. Those drives are arrays of something like 64 devices.
Re: (Score:3)
It's the size of a single double-wide PCI card. Okay, scratch that, it *is* a single double-wide PCIE card. That counts as a single device. Just like how if you put a bunch of hard drive platters behind a common interface within a standard-size hard drive shell, it counts as one hard drive.
Re: (Score:2)
They probably cheated by putting more than 96 NAND dies in their device. 96 NAND dies should be enough for anyone. (sorry)
Re: (Score:2)
That's not exactly a single chip.
Re: (Score:2)
Just add more (Score:2)
Re:Just add more (Score:5, Informative)
It costs money to stack. At a much higher rate than it does to scale. Or at least that has been the case. It will be a significant hit to the industry when they can no longer count on device scaling to help bring up density, and get forced to wire multiple chips in ever expanding arrays.
Did Anobit have anything to do with this? (Score:2)
I only ask because Apple is the largest flash customer/reseller in the world and they just bought this company
Flash retention times (Score:2)
Along with error rates, what will happen to retention times as the cell size shrinks?
Supposedly, flash memories have expected retention times as short as 5-10 years or so (if not refreshed by re-writing), thanks to gradual leakage of the trapped charges they use to record data; this value is expected to drop as flash cells get smaller. I've had gadgets whose firmware mysteriously become corrupted after sitting around for a few years, and sometimes they could be revived by re-flashing them -- I sometimes wo
Re: (Score:3)
Re: (Score:3)
Retention time in 2003-time-frame flash is tens of years. Retention time for the latest 25nm flash is measured at one year. Much less if you wear it out. Your 8MB SD card likely hasn't had the level of cycling needed to see reduced data life.
Do they still use unbalanced conductors in ... (Score:2)
... microelectronics fabrication, making them vulnerable to inductive effects?
NAND Successor (Score:2)
I was kind of hoping we'd have something better than NAND Flash within 5 or so years. Maybe something using memristors? NAND is just too expensive to be useful. Prices haven't dropped in a couple years.
Re:NAND Successor (Score:4, Informative)
Prices are now down to about $1.50/GB for standard 2.5" SSDs. And you can sometimes find them for $1.25/GB. That's lower then the $2.50-$3.00 of 18-24 months ago.
Sure, it's expensive compared to the $0.10/GB of bulk storage like 1/2/4TB drives, but when you compare it to things like 10k RPM SATA/SAS and 15k SAS (about $1/GB) it starts to not look so expensive. The only things that make me nervous about them is that SSDs still have some controller issues and it's a younger technology compared to traditional hard drives.
At $1.50/GB, that means you can purchase a 120GB SSD for about $180. For a lot of people, that's big enough and cheap enough in exchange for vastly improved performance. And if you can keep the users from storing stuff locally, you could go with one of the 64/80GB units which are in the $100-$125 range.
I've converted a few users over to SSD over the past 2 years. It's been worth the money every time. The machines are far more responsive to user input, they don't sit there and spin, and it generally means that the CPU starts being the bottleneck again. Not all of these are power users, either.
I paid about $1.75/GB for my 250GB SSD. Do I wish it was bigger? Sometimes. But it turned a 4-year old laptop from something that I hated using due to the slowness of the old 500GB 5400 RPM hard drive into something that is fast and responsive. For work it made me much more productive.
two types of storage (Score:2)
Type one: bandwidth sensitive. OS files, application files, cached application data, etc.
Type two: bandwidth insensitive, e.g. streamed. E.g. Video, audio.
Store type one on a SSD. Store type two on a higher-capacity magnetic drive. How likely is type one data expected to grow? Perhaps not that fast. My home machine is an ~8 year old Dell laptop with a 60G disk. It's not even half full. That includes an OS, browser, Office suite, and a few other applications.
What about PCM and MTJ? (Score:2)
There's been tons of research on alternative technologies, including phase-change memory (http://en.wikipedia.org/wiki/Phase-change_memory) and magnetic tunneling junction (http://drl.ee.ucla.edu/index.php?page=research&function=sttram) memory. Obviously commercializing them is expensive, but some progress has already been made there. I'm sure other competing technologies will be developed in that time as well.
Easy solution to the scalabilty (Score:2)
only use the ssd partitions where it makes sense. no reason to install the whole OS to the ssd. just mount /tmp or /var/lib/mysql on an ssd slice. use plain old sata raid for the rest.
Several fundamental flaws in their assumptions (Score:4, Interesting)
While they discuss individual SSDs, modern flash storage arrays ( http://www.violin-memory.com/products/6000-flash-memory-array/ [violin-memory.com] ) can hide all the write latency and its effects on read latency. When you start talking about 16TB SSDs the same techniques can be used.
As far as bandwidth and IOPs, they use a 4K/8K write size for MLC/TLC, but MLC already exists with 8K pages, as well as having the ability to write more than one plane at once, which doubles the write bandwidth. Double the page size again and you double the BW.
Now bigger page sizes only help on the reads if you can use more than a single user read worth of data in the page, which might be possible depending on what the system knows about access patterns. But without making assumptions about the ability to store data together that's likely to be read together, garbage collection, which can wide up reading more bytes than the user does, can use most of the data in a page.
So there are factors of 2X, 4X maybe 8X in performance that the paper misses out on.
As far as density, it is not necessary to go to smaller features to get more bits per chip by using 3D techniques such as Toshiba's P-BiCS (Pipe-shaped Bit Cost Scalable) MLC NAND which allow vertical stacking which increases density without using smaller features with their worse performance and lifetime.
The group at UCSD that authored this has done some nice work so I don't mean to be too negative, but they are trying to predict too far from a limited and faulty set of assumptions which unfortunately negates much of the validity of this paper.
jon
p.s. in the interests of full disclosure, I make the arrays in the first link :)
Re:HDDs for the win! (Score:5, Funny)
Yes, please send your SSDs to me for disposal, thanks.
Re: (Score:3)
No, no one remembers it. It's an invention of fantasy, not memory.
http://en.wikiquote.org/wiki/Bill_Gates#Misattributed [wikiquote.org]
Re: (Score:3)
The funny thing is that this "myth" was well established long before he published any sort of rebuttal.
This is ancient history. It happened decades ago and finding evidence now would be difficult even if you knew where to look. Chances are that any such corroboration faded away by the time that rebuttal was published.
Re: (Score:2)
Because, maybe, one of those children will innovate, and solve this problem?
Re: (Score:2)
I will be happy (but not satisfied) when just a mere 1TB SSD does not cost the same as a house or a (decent) car.