[PCWorld]

Post your comments for SSDs Have 'Bleak' Future, Researchers Say here

[RGOM]

Good article, interesting. Seems to point to continued research in a number of different dirctions: can the MLC or TLC manufacturing process be improved to increase reliability? Are there alternative "disruptive" technologies out there that need to be brought out of mothballs and developed? Do we really need 1 TB SSD's at the customer level -- can folks develop neat, easy applications for archiving? (For almost 3 years I've been getting by with 64G and 128G SSDs by use of auxiliary storage -- writing documents and photos and other archival stuff to SSDs or downloading e.g. photos to disk-drive backups). The idea would be to improve on the hybrid SSD-magnetic disk drive pairing. . . e.g. pair an e.g. 64 gig SLC SSD with an on-board archival disk-memory, and when the SSD wears out and bogs down you just replace it. Cleverness around the use of "docking stations" and archiving to the dock, might help, too.

[agrippa]

"SLC flash has the highest reliability and resiliency, with 50,000 to 100,000 program/erase (P/E) cycles. MLC NAND can sustain 5,000 to 10,000 erase cycles. TLC NAND has the lowest endurance, with 1,000 P/E cycles to as few as 500 P/E cycles, the researchers found."

So what does that mean for a average user? how many years would that be? any ballpark figures?

[Nuke61]

This article, in conjunction with another one I read, now make it crystal clear why Apple purchased Anobit. I thought Anobit was just another memory controller company, and Apple wanted to have their own proprietary controller, but it looks like Anobit's technology is addressing, head-on, the problems listed in this SSD article.

Quote

At a demonstration at the Flash Memory Summit Anobit showed the difference between its controller and a controller with 24-bit error correction, which is understood to be pretty advanced. The demonstration used a NAND flash chip with pretty horrid behavior – a 3-bit per cell chip that started to lose data at fewer than 300 erase-write cycles. Anobit wrote and erased each block in the flash a different number of times – the first block got one erase-write, the second two, and so on.

The demonstration stepped one-by-one through these blocks, all of which contained the same photograph. At around 500 erase-writes the standard 24-bit error correction stopped being able to recover the picture, but the Anobit-corrected picture stayed intact for more than an order of magnitude more cycles, indicating that this algorithm could get far more life out of a flash chip than could standard algorithms.

Why is this important to Apple? Well, as time marches on and NAND flash chips migrate to increasingly aggressive processes to reduce costs ever further, the chips become increasingly more error prone and harder to control. Flash users will be forced to use continually improving error correction schemes to be able to get their products to work at all.
http://www.zdnet.com...quisition/11870

Basically, this article talks about MLC NAND being able to do 5-10K cycles, but the Anobit technology takes MLC flash and enables it to reliably do 50K cycles or more, so you get near-enterprise level reliability using consumer grade (priced) parts.

[ReadandShare]

An interesting article made dramatically useless by sheer absence of any perspective for users!

[waldojim]

agrippa, on 16 February 2012 - 05:46 PM, said:

"SLC flash has the highest reliability and resiliency, with 50,000 to 100,000 program/erase (P/E) cycles. MLC NAND can sustain 5,000 to 10,000 erase cycles. TLC NAND has the lowest endurance, with 1,000 P/E cycles to as few as 500 P/E cycles, the researchers found."

So what does that mean for a average user? how many years would that be? any ballpark figures?

The latest drives have about a 3 year life expectancy.

[Evildave]

Another big dependency is how it's used, for that lifespan. If you rarely (if ever) modify the contents, it will last 'forever'.

So an OS+Application drive will last a long time, as long as the temp/scratch/swap/log/etc. files are written 'elsewhere'. Of course, earlier versions of windoze that would let apps write into the 'Program Files' folder would break that assumption nastily. But latter days versions behave OK in this regard.

Another thing to keep in mind: DO NOT 'fill up' a SSD. Most of the controllers/operating systems are smart enough to cycle through 'free' blocks, writing changes into all of them in turn, rather than the same ones over and over. Fewer free blocks means more punishment on the same 'free' cells. Journaling file systems write new copies of files, rather than overwrite them. There's still indexing to keep up, though, but that can be reconstructed readily, so 'lazy' commits can keep the writes down on that. And with a huge RAM cache against the SSD and guaranteed backup power (battery in a portable device), lazy commits can be done all around with no data loss. Just 'sleep' the machine well ahead of battery depletion, with enough capacity left to guarantee all commits will be completed before it fully sleeps.

Also, with 'lots' of RAM, the need to 'swap' and write a lot of temporary crap to the hard disk is much reduced for MOST applications. 16GB of RAM is as cheap as only 1GB was, just a few years back.

Modern SSD flash has 'millions' of write cycles, not tens of thousands, but that's mostly due to the clever management of writing, and error correction. This comes at a 'cost', but it's a cost people seem willing to bear, not to have a portable device whir and click and vibrate in their hands, and eat lots of power all the time to keep something spinning and moving around inside, while making it ultra-fragile versus common daily mechanical abuse, compared to solid state storage.

Also, MTBF/MTTF for a hard drive is different from its life expectancy. You'd still need to keep replacing a mechanical hard drive every few years to get the same component level 'reliability' as predicted for a new mechanical drive. A six year old mechanical hard drive is a ticking time bomb for data loss. How many times would you have really written to the same sectors on that mechanical drive in five years or so?

The most central thing that the students who did this study fail to consider is that other technologies grow apace with the capacity. Better fabrication techniques. Better error correction. More clever techniques all around. DIFFERENT SSD memory cells.

In five or six years, the cost and capacity of SSDs will be the same as any notebook sized mechanical hard drives, and within a couple of years of that, the relative faults and merits of the technologies won't particularly matter anymore: Everyone will use SSDs. The copper and rare earths will be wanted in a big way for motor vehicles, meaning those manufacturing materials will keep going up in cost, so the parts in consumer grade mechanical hard disks will probably get worse and less reliable, not better, as they continue to seek ways to cut costs and keep competing. Not that cleverness won't help the mechanical drive makers... but they're the ones already up against the wall with the density of magnetic domains in their storage, while SSD is still only scratching the surface of their potential data storage density.

Whatever storage media you use, you don't get away from needing to back up your data. It has never been optional, and it never will be (at least in our lifetimes).

[kronoscornelius]

Really good article. @Nuke61 IBM has also done some work on improving SSD reliability by doing something that is similar to a Raid-3 at the chip level.

@evildave, the student does well in not speculating about the future, science is about measuring the and extrapolating real data. Maybe a market analyst can help you in considering how multiple technologies will shape the industry.

From my shopping SSD shopping experience, I know I like my SSD's to be "cache" devices where I keep the data that needs to loaded fast and not written too often (favorite programs , OS), and I use the HDD for my data in Raid-0 partitions. This research advices me to keep doing the same for the next 10 years.

[LiveBrianD]

waldojim, on 16 February 2012 - 07:43 PM, said:

agrippa, on 16 February 2012 - 05:46 PM, said:

"SLC flash has the highest reliability and resiliency, with 50,000 to 100,000 program/erase (P/E) cycles. MLC NAND can sustain 5,000 to 10,000 erase cycles. TLC NAND has the lowest endurance, with 1,000 P/E cycles to as few as 500 P/E cycles, the researchers found."

So what does that mean for a average user? how many years would that be? any ballpark figures?

The latest drives have about a 3 year life expectancy.

What would you say for a near-enterprise drive like a WD Caviar Black? I mean, they do offer a 5 year warranty on those.

This post has been edited by LiveBrianD: 16 February 2012 - 09:59 PM

[karthiq]

Well then i guess SSDs will be succeeded by phase change memory.

[waldojim]

LiveBrianD, on 16 February 2012 - 09:56 PM, said:

waldojim, on 16 February 2012 - 07:43 PM, said:

agrippa, on 16 February 2012 - 05:46 PM, said:

"SLC flash has the highest reliability and resiliency, with 50,000 to 100,000 program/erase (P/E) cycles. MLC NAND can sustain 5,000 to 10,000 erase cycles. TLC NAND has the lowest endurance, with 1,000 P/E cycles to as few as 500 P/E cycles, the researchers found."

So what does that mean for a average user? how many years would that be? any ballpark figures?

The latest drives have about a 3 year life expectancy.

What would you say for a near-enterprise drive like a WD Caviar Black? I mean, they do offer a 5 year warranty on those.

I have seen some that last for 10 years or more. Shoot all of my 10K scsi drives still work. Those suckers are over 10 years old.

[Evildave]

Plug in a dusty drive out of the closet (that hasn't failed, YET), it'll probably work 50 years later, if you can find the cabling/interface for it. Especially if it's just been sitting in a closet for most of that time.

Same for SSD, for that matter.

The problem is, after a few years of daily use, all bets are off on the 'reliability'. It could work for another few years.... It could fail any time.

All predictions of long term durability are probabilistic. A 'million hours MTBF' means if you take a hundred drives and run them for ten months, ONE will probably fail (give or take). It DOES NOT mean a hard drive will last a hundred years.

But after a few years, the failure rate starts climbing. Pretty soon one is failing every month or two. Hopefully some SMART status gets noticed before it goes, but this isn't always the case. You can still keep your remaining drives running, replacing them piecemeal, but you're going to be busier all the time replacing drives and restoring backups. In ten years' time, you could still have some hardy survivors from the original batch... but you probably would've replaced them all long before then, just to be able to do something OTHER than replace failing hard drives for a living.

[waldojim]

Hey dave, before you ASSume anything, stop to think for a change.

Those hard drives are what i use in my machines that do little to nothing. Even though they run 24/7. Not everyone uses a closet case to make thier assements of equipment lifespans.

[Kazmatron]

I actually end up replacing my HDDs to get more capacity before they fail. I've only ever had 1 failed HDD, and it was a DOA.

[Evildave]

But a hard drive doing 'little to nothing' IS LITERALLY the 'closet case'.

SSD is just as durable as mechanical in that case. Moreso: An earthquake (or klutz) won't make them crash mechanical heads into spinning platters.

Everything about how 'doomed' SSD is, is pure, unadulterated FUD.


Yeah, you're probably gonna update faster than the SSD fails, just as you update faster than HDD fails.

I had a really expensive 1GB HDD that lasted a couple of months... and then died. Bloody bathtub curve. Paid $800 for it, back sometime around 20 years ago. SCSI, high reliability, 'really fast' (for the day), etc. It's the one that taught me firmly about 'Backup, backup, backup!'. Because I did lose some stuff to that disaster. Actual 'work', if you can believe it. Getting a 'replacement' was absolutely no solace.

[waldojim]

Evildave, on 17 February 2012 - 10:56 AM, said:

But a hard drive doing 'little to nothing' IS LITERALLY the 'closet case'.

No, that is not the same thing. Never has been. Little to nothing does not mean nothing. If the drive is running yet not writing all day long, it is STILL running. You apparently have no concept of this. This is actually one place mechanical drives are the weakest. The actual surface of the drive is normally not the problem.

Quote

SSD is just as durable as mechanical in that case. Moreso: An earthquake (or klutz) won't make them crash mechanical heads into spinning platters.

Everything about how 'doomed' SSD is, is pure, unadulterated FUD.


Yeah, you're probably gonna update faster than the SSD fails, just as you update faster than HDD fails.

I had a really expensive 1GB HDD that lasted a couple of months... and then died. Bloody bathtub curve. Paid $800 for it, back sometime around 20 years ago. SCSI, high reliability, 'really fast' (for the day), etc. It's the one that taught me firmly about 'Backup, backup, backup!'. Because I did lose some stuff to that disaster. Actual 'work', if you can believe it. Getting a 'replacement' was absolutely no solace.

FUD? Failure of SSDs? Yes, the new drives have a very short lifespan. All current data points to premature failure being a problem with the newest flash types.

In either case, backups are important. Glad to hear you actually learned something regarding backups.

[Nuke61]

Admittedly, this is almost ancient in computer years, having been written back on 2007, but Google did a study of hard drive failures in their facilities and found some interesting things.

MTBF cannot be used to estimate the life of a single drive. If a drive has a spec'd MTBF of 300K hours, that means that for a large population of drives, at 300K hours fully HALF of them will have failed. Furthermore, the Google study and another one conducted by Schroeder and Gibson at CMU concluded that the real world MTBF was roughly 1/2 of the advertised MTBF.

If a drive is pushed hard (lots of reads/writes) in its infancy, it will suffer a higher failure rate than drives that are lightly used. However, once past that early in life period, it doesn't seem to matter whether a drive has lots of reads/writes or just spins.

RAID 5 for backups? They concluded it was not such a good idea. Google concluded it was much better to simply have a copy of the data on 3 different hard drives. High temperatures are bad for drives, right? Only if the drive is old, again, per Google data.

In short, replace your drives if their are over 3 years old.
Don't rely on RAID for backup. Have the same data in 3 different locations.

http://storagemojo.c...ure-experience/

http://storagemojo.c...disks-is-wrong/

[Evildave]

waldojim, on 17 February 2012 - 11:43 AM, said:

In either case, backups are important. Glad to hear you actually learned something regarding backups.

Only a few months back, you were arguing your 'RAID' was more than adequate, and too much trouble to back up. I take it you finally figured out how to plug an external drive in, start up rsync, and back it up while you watch TV, then?
http://forums.pcworl...security-freak/

It's good that you have finally waken up and admitted the value of backups, something which I have repeatedly and patiently preached to you.

This post has been edited by Evildave: 17 February 2012 - 04:27 PM

[waldojim]

Evildave, on 17 February 2012 - 04:20 PM, said:

waldojim, on 17 February 2012 - 11:43 AM, said:

In either case, backups are important. Glad to hear you actually learned something regarding backups.

Only a few months back, you were arguing your 'RAID' was more than adequate, and too much trouble to back up. I take it you finally figured out how to plug an external drive in, start up rsync, and back it up while you watch TV, then?
http://forums.pcworl...security-freak/

It's good that you have finally waken up and admitted the value of backups, something which I have repeatedly and patiently preached to you.

For my needs, RAID works just fine. I don't go around accidentally deleting things. Not to mention that the entire contents are that of movies I have bought, paid for, and can easily re-rip. Well - that and a few family photos, which end up burned and mailed off to family that has very limited download bandwidth.

EDIT: You do preach. Thus the reason I fight you at every turn. Because you feel the NEED to be right with everything - even when you are blatantly WRONG.

This post has been edited by waldojim: 17 February 2012 - 05:56 PM

[LiveBrianD]

waldojim, on 17 February 2012 - 05:55 PM, said:

For my needs, RAID works just fine. I don't go around accidentally deleting things. Not to mention that the entire contents are that of movies I have bought, paid for, and can easily re-rip. Well - that and a few family photos, which end up burned and mailed off to family that has very limited download bandwidth.

EDIT: You do preach. Thus the reason I fight you at every turn. Because you feel the NEED to be right with everything - even when you are blatantly WRONG.

If the array did fail or something (that happened to my mom once, raid 1), and you lost all the data on it, it would take quite a while to re-rip all those movies. Honestly, isn't it just easier to backup everything, so it's just a simply matter of copying all the files back?