SSD Endurance - the Big Lie

Lately, I've noticed that many people seem to have a warped and exaggerated idea as to how long their SSDs will last - many proclaiming with enthusiasm that their SSD shall keep running for decades or even centuries to come.

I decided to write a post about this. I'd like to prove false the biggest lie the manufacturers would have you believe:

(SIZE OF SSD) * (Endurance Cycles) = Total data written to SSD before failure

This is utterly, truly, and inexcusably FALSE.

The problem with an SSD is, data is written in blocks. A block may be 256KB: 256 * 1000 * 8 binary digits. To change even ONE of these digits, you must rewrite the ENTIRE block. That is to say, your OS sees 1 bit being written, but the SSD wear is equivalent to 256KB being written: a 2.048 MILLION fold difference.

The aforementioned scenario is the BEST CASE SCENARIO (and what do you think the manufacturers would have used? geez.). That ONLY occurs when you are writing data in HUGE chunks (120GB file to a 120GB SSD). In that case, yes, you will be able to write 120GB * 10000 = 1200TB data to the drive before failure.

In real OS use, the SSD will be worn extensively by the OS. Suppose your OS writes 2GB of data to the SSD per day (a very, very, very conservative estimate). You might think that since your SSD has 120GB capacity, it should only give you 0.01667 cycles wear. That is NOT true. Many of the OS writes are tiny little operations:

• Windows Live Messenger performs about 20,000 write operations upon log in to various logs and caches
• Firefox performs about 10,000 writes for opening 3 web pages and one five minute youtube video

These figures are gathered using procmon.exe released by Microsoft.

What does this mean? It means that the little 2GB you wrote to your 120GB SSD might actually cause an increase of 3 or 4 cycles in your average SSD wear cycle count! This is hundreds, even thousands of times greater than many expect.

All of this is based on my personal research and experience. I have a 120GB Vertex that I meticulously used. I have moved every imaginable cache and log file off the drive and onto a RAMDisk. On the other hand, another 120GB Vertex in another system with an almost identical usage profile suffered about 2x the amount of wear of mine in the same period of time. This suggests that what I am doing is making a huge difference.

Even so, all monitoring programs (CDM, IndilinxSSDStatus, SSDLife) are reporting 120 average wear cycles from 160GB written. This once again proves my point about the real wear.

If you have any additional info / technical data, please post it in this thread.

Here is a most excellent read about actual SSD semiconductor technology, and wear. http://www.lostcircuits.com/mambo//index.php?option=com_content&task=view&id=69&Itemid=1

**** Update ****

NEM asked whether I could post a guide to relocating these folders. Here's how:


1) Grab the free DATARAM RAMDisk from http://memory.dataram.com/products-and-services/software/ramdisk
2) Install the RAMDISK with default settings, and startup the RAMDisk. I have a 400MB RAMDisk on 4GB total memory.
3) Grab procmon.exe from http://technet.microsoft.com/en-us/sysinternals/bb896645.aspx
4) Run procmon.exe in the background while you do your daily tasks. Then, click on Tools -> File Summary, and sort by Writes. Note the locations to which absurd amounts of writes are taking place, and decide whether it's a cache that you don't really need. Things like temp files are unnecessary. Things like cookie lists and bookmarks, though.. well, you decide.
5) Close all programs.
6) Go to the location with the FOLDER containing the frequently written to data. Copy the contents of that folder to your RAMDisk and leave under root directory. Delete the whole original folder.
7) Type in cmd into the run bar in the start menu, right click the result that shows up and select "Run as Administrator". Use the following command: "mklink /d (COMPLETE_PATH_OF_ORIGINAL_FOLDER) (RAMDISK_DRIVE_LETTER):"
8) Start the program that was making those writes, and use procmon.exe to verify that the writes are now going to the RAMDisk.

Tips:

- If it says "cannot delete folder" in step 6, try closing more programs, or starting up in safe mode, or make a new user account in Windows and try doing it from a different log in. Make sure you log out of the original account, though.
- Under system variables (right click My Computer, go to Properties, Advanced System Settings, Environment and Variables) modify all entries that mention "Temp" or "TMP" and make them point directly to the root directory of your RAMDisk.

Notorious programs for writing too much:
- Skype. 30k writes upon startup for me, and another 20k over a few calls (what the hell is it doing, logging my voice and sending to CIA???)
- Windows Live Messenger. 20k writes upon login to various logs.
- Firefox. HUGE number of writes for opening one page (1-2k for opening NBR)
- more to be added.

Anything else to be added soon.

 

There's also the issue of write amplification which must be factored in when trying to determine theoretical SSD life. So, one is never write just say, 100MB, when writing a 100MB file. The true size of the write will be several orders of magnitude larger due to write amplification.

Here's a fairly detailed wiki to explain write amplification: Write amplification - Wikipedia, the free encyclopedia

 

In your opinion, how long will an average SSD last?

I'm keeping mine until the 3 year warranty ends.

 

With my usage pattern (browse net, play a few games, talk on MSN, Skype, etc) it appears that my SSD will last about 8 years. However, my SSD:

- has 10000 specified P/E cycles. Many of the newer SSDs based on a smaller manufacturing process only have about 5000.

- is optimized. I have created about 20 folder symbolic links to move stuff off the SSD. When I monitor my writes, I seem to have redirected about 80% of the small writes OFF my SSD.

 

Doesn't that defeat the whole purpose though? Most buy an SSD as a HDD replacement. I think what you're doing is a good indicator of reliability (or lack of confidence in its reliability), atleast to me anyway. Nobody should have to do that to extend the life of their HDD.

 

Hmm. Well, it's so so. Moving cache and log files off has minimal impact on my system for two reasons. The data is stored on a RAMDisk, which has about 7GB/s read and write speeds. Secondly, when I restart, I don't expect to use this kind of data from the last session again anyway.

 

I think you misunderstand me. I agree with you on your assessment that SSD's are not reliable. I was simply stating that nobody should have to go through such great lengths to increase the reliability of their hard drive, and that speaks to me that atleast some out there are not confident in SSD's reliability.

 

Oh ah I see I misunderstood what you meant. I thought you were saying that moving stuff off the SSD defeats the purpose of having an SSD for the speed aspect. I forgot that people buy SSDs for reliability too, lol but it's definitely not write endurance.

 

hey min, would you be interested in posting a small guide of how you're redirecting these writes? I'd really like to do the same and I'm sure a lot of people would find it useful.

 

I never even thought about it, I went for the speed, and most importantly, the battery/power savings.

Four to eight years sounds horrible, but to be fair, I do not think I have ever relied on a hard drive for longer than that. By that time it has been relegated to an external for fun.

I suspect in four years I will either have a better laptop, or be buying one of those dirt cheap 1tb SSD deals on Black Friday .

 

I think that's what they're counting on though. Companies are counting on people buying new systems every one or two years, but there's a very small percentage that keep their systems around for years as they don't need the latest and greatest.
My brother told me my mom gave him her old emachines from 1998, and it's still chugging along just fine, including the hard drive.

 

4-8 years is fine for me, I have never had a hard drive longer then that.

By the time the drives fails from to much use ill probably want newer technology anyways.

 

Hi, update made to main post.

As for 4-8 years: problem is, that's an estimate on ... I guess when the drive becomes read-only. The performance degradation should probably become noticeable way, way before then.

 

Yes, but these people do not have SSD, and will not until it becomes pretty industry standard. I am sort of surprised how bad SSD prices still are from manufacturers, where most people are getting their laptops. HP was pretty much the only one with a "reasonable" upgrade cost (and I think that was because they were bumping up a free upgrade to a better drive already).

Dell still wants something like $500 for a single SSD drive upgrade! (Keep in mind, not only are SSD's a fraction of that price, they are already factoring in some, albeit small, cost for the base hard drive. I am talking +$500!

 

4-8 years... sounds quite average. Most HDD's are supposed to have an average life of 5 years lol plenty of peoples last longer (I'm on year 6 of my 110GB hdd) and some don't last nearly as long...

really if it lasts 5-6 years that's plenty... and as software optimizes for SSD's (firefox is planning on it) they'll last longer.

edit: and just as a hard drive you'll see performance degredation. Really this sounds pretty standard.

 

There is no performance degradation with a magnetic drive.

 

When you start reallocating bad sectors I believe there is a performance penalty...?

I've had a fair number of old drives with high reallocation counts in SMART, and the HDTune curves are all over the place...

 

That's no different than companies that are still charging $170 to "upgrade" to 4GB RAM (from 2GB), when you can buy that for $80 or so and install it yourself.

 

There sure is...ever heard of bad sectors and S.M.A.R.T?

 

I guess the question then is will the SSD degrade past the point of its' purpose? If it ever becomes as slow or worse than 7200 mech drives, then yeah it's a serious flaw. But if it starts at say, 250mb/s writes and ends its' life doing only 120mb/s writes, you are still pretty far ahead of a mech drive and probably got good use out of your ssd.

And yeah OP, it is a big lie, but realistically they can get away with it for now. SSD's are still only used by maybe the top 2% of computer users because of cost, and with tech changing every 12-18 months most people who use them now will have long since replaced their ssd by the time the one you have now critically fails.

 

no SSD vendor claims their life the way you present it. Your's is the best case scenario.

For most SSD I can't even find such a number(how many bytes). The only thing I can find is Intel which have these documented. Though they haven't said what pattern they were used to reach the mentioned numbers. For my x25M 80G, it is quoted to be 7TB(but in real life, it should be way more than that).

 

Can't say I've had any bad sectors in the last five years or SMART (stupid) problems ever.

But regardless, none of that is part of its normal operation, unlike with an SSD.

 

Well, here is the problem. Many vendors don't calculate the number out for you, but they hint at the aforementioned method of calculations. That's why so many consumers are tricked into thinking this.

Even if they quote 7TB: it really depends on what you do with the drive. I've reinstalled my system a few times since I got the SSD (which didn't affect the count at all). I check the amount written versus cycles worn almost daily. Usually I find that it's about 1 cycle for every 0.5-1GB written for normal OS use. That gives something like 5-10TB writing before breakdown, which means Intel's claim is actually not that far out of the ballpark. Again, this is all dependent on how you use the SSD, as I described. Even your Intel SSD could potentially be destroyed after writing a few GB to it if the code is malicious.

But if you go over to the OCZ forum, they are discussing the fact that newer processes produce NAND memory that might only have 4000 or 5000 P/E cycles. That just dramatically reduced the lifetime to maybe even 2 years. There are people showing drives that have been used for one year and are at thousands of cycles written. That is pretty bad. Look at this:

Support Question OCZ Vertex Turbo FW 1.6 - Health query

Particularly, Support Question OCZ Vertex Turbo FW 1.6 - Health query

See the difference in the ratio of cycles worn / GB's written? Especially the Vertex Turbo 120GB - 1680/5000 cycles used up by writing 760GB. Kind of scary. Definitely not what OCZ claims.

 

Thanks min, +rep

 

i think 4-8 years is acceptable.
personally id be more interested to find out how fast this degradation occurs during normal use.
i see a lot posts with ssd owners trying to 'baby' their ssd's by turning off page filing, moving files and such to external drives, not using torrents, etc. if you dont baby your ssd, how long would you see before it degrades in performance and how long till it 'dies'.

 

+rep awesome info man. Bookmarked this so 5 years down the road when they are more reliable and cheap i'll do this ^^

 

Well, essentially all the crap like logs that have entries like "2010-12-01-12:51:11 xxxxx Loaded" or temporary internet caches are moved to RAMDisk, so it's automatically lost when shutdown. I think this limits damage to SSDs significantly.

Again, like I said, 4-8 years is depending on how you use the SSD, the specific NAND used, and I guess what you define as "not usable anymore". In the OCZ forum link I sent, there are SSDs that are 1/3 gone after maybe just a year of operations, and I'm thinking that 1/3 gone means already a significant portion of cells having become read-only so there is already a noticeable loss in performance.

DCMAKER, I got a bit of bad news. It turns out that I got lucky with my SSD. The exact same model is available in two types, with the earlier type having NAND produced using a larger manufacturing process than the newer one. The newer process costs less, but only has 5000 P/E cycles compared to the earlier 10000. I don't like the way that trend is going...

EDIT: From my research, as well as this thread: http://www.ocztechnologyforum.com/forum/showthread.php?78174-PE-cycles-or-write-cycles

it appears that almost all new Sandforce based SSDs use 34nm process NAND with 5,000 cycles only. This is half of the 10K cycles of the 40nm NAND.

 

1) Is this all theory, or does anybody actually have a documented case where an SSD has used up all of its write cycles, and reverted to a read-only mode?

2) Assuming an average of 5 years - does anybody here actually keep a drive longer than 5 years? I have replaced every drive I have ever owned within 5 years to upgrade speed, upgrade capacity, or replace a dead drive.

 

I admit, I have not personally seen what happens with a drive that uses all of the cycles. However, since manufacturers themselves claim that a single NAND gate can only sustain at most 10K cycles on average, I think it's pretty alarming.

Here's the thing, if you're only using maybe 1/2 of the drive capacity, the drive could potentially juggle data around so you don't notice until more than half of the drive is dead.

On the OCZ forum, I recall reading a moderator stating that the drive will lock itself into read-only mode when the limit is reached. I'm not sure whether this is true, or whether it applies to other drives also.

The point of my post is mostly to draw attention to the fact that many people have a far too high expectation as to how long an SSD will last. The manufacturer's claim of "capable of 10 years of operation at 10GB written per day" is about as useful as a brake pad manufacturer saying "capable of 10 years of operation, driving 10 hours a day". And then it turns out that the claim implies that you drive for 10 hours on an empty highway and brake once at the end of that journey. It may have nothing to do with real life usage.

As for keeping a drive for 5 years, here's what I'm saying, though. We spend a large premium on SSDs expecting top notch behavior, so a 20% performance degradation after 1 year is kind of unacceptable, right?

 

just loosing these logg is a bad idea as you need them when trying to figure waht went wrong after the event so just loosing them is a bad idea so having them moved to a ram disk and then writen on the disk only one time when you shut down bring the best of both world

 

DATARAM allows for that, actually. So all you have to do is enable it. It'll make startup and shutdown slightly slower, though.

Although, in my years of using a personal computer, I've rarely ever looked at logs to find out what went wrong. Normally I can just Google the symptoms.

 

min22009 thank you so much for the write up. I too am perplexed at why people think SSD's are more reliable, when they are but not by enough to warrant the price difference as a double or quadruple drive mirror RAID configuration would provide far more security.

However Min2209, If I am not mistaken the advantage of a SSD over a HDD is in terms of data security is that even if the drive fails, the only issue is that writing can no longer be done and the flash memory should be permanently set. Reading of the data is still functional, thus you can extract that information. Companies who rely on security of data, may use SSD because of that aspect, for a company the data stored is far more crucial then the cost of the drive it self.

Correct me if I am wrong, my research in SSD technology isn't up to what I would like it to be.

EDIT: From my understanding of SSD. Most users will experience a 4 year~ life expectancy for their SSD (maybe less). It can be stretched upwards to 6-8 years if heavy modifications programs, some adjustments to the OS, and certain caches/repeatedly accessed/rewritten files.

 

Because people don't buy SSD's for reliability. Things like battery life and greater resistance to mechanical and thermal shock are nice-to-have. But when it comes down it, an SSD buyer wants performance.

Desktop users want the fast OS boot times, application load times, and multitasking smoothness. Enterprises want the high I/O's that SSD's give for I/O intensive tasks like databases.

 

But reliability comes with the package. No moving parts means higher reliability, as well as less noise, considerably less heat.

 

Yes, reliability comes with the package. But it is a side benefit, and not a primary benefit. Anyone looking for reliability and data security will still use RAID arrays and create backups / drive images.

An SSD may reduce the chances of a drive failure because of no moving parts, but nobody looking for reliability would be foolish enough to skip all of that, buy a single SSD, and call it a day.

 

Well if you were constantly on the go, business meetings, coffee shop, airplane travel and needed a laptop, you would NOT consider an SSD? An SSD would be the first thing on my mind.

 

I think we're talking about the same thing in two different ways. I am pretty sure that we can both agree that:

1) An SSD is more reliable than a hard drive, because it does not have moving parts.

2) A single SSD, by itself, is not a replacement for proper disaster recovery practices like RAID and backups.

Sound about right?

 

If I were constantly traveling I would do a SSD/HDD combination on a workstation. Most HDD's have fairly decent protection against movements on off axis angles. Bumps is another story, which is why I would do SSD for operating system/critical files and HDD for storage/programs. But to be honest most businesses utilize other means to transport high critical files such as encrypted flash memory devices. Or network accessed storage utilities hosted on physical servers.

Yeah SSD do save on battery, yes they run cooler however my argument is on the security pro's and cons.

Most companies run servers/desktop computers for important matters in which case RAID mechanical hard drives are king, and so are high powered high RPM fans for the servers.
Even critical data is sometime stored on extremely encrpyted high durability flash memory devices for mobility, not a mobile workstation (laptop).

 

Oh of course, you can never just rely on a single drive, albeit an SSD or not without backing up your data.

Of course 90% of my computers at home are for gaming so no biggie if a hard drive dies really.

 

What will happen to the new Macs that will apparently have their SSDs soldered straight onto the board?

Edit: Nevermind, apparently that was just a rumor, it's removable.

 

Yes the SSD in a Toshiba mini PCI-E form factor SSD.

The RAM is soldered onto the motherboard and must be upgraded through Apple or desoldering it yourself and risk voiding your warranty.



Yellow highlights is the RAM.

 

ypu got to be kidding me ... ram soldered ...... this gottabe the dumbest thing i've heard what's next you wont be able to remove your batterie to prevent it from over charging

o wait .....

 

Not too much of a big deal.

Most Mac owners I know will just buy a completely new machine if any one component goes bad. Having a dead SSD doesn't change that equation.

 

You can remove the battery...you just need to remove 8-10 thin Phillips (depends if you have MBP, Macbook, and Air has this same design) to take off the bottom cover.

Apple was smart designing this, you are forced to buy and pay Apple prices up front, no upgrading if you don't know what you are doing.

 

And Apple continues to feed the stereotype of Mac users...

 

From a "business" aspect as discussed I purchased the SSDs for many of the reasons mentioned - speed, multi-tasking improvement, quiet operations and lack of moving parts as the laptop is on several times while moving around.
Have used it as a "normal hard drive", part of my usage includes about 10GB of data being updated every two weeks. Have done so for the last 6+ months. SMART data shows media wear out at 99. If that continues I'm fine with it. Bought the drive for the benefits - not to worry over what my usage patterns might do to the hard drive.
Usually ran through mechanical hard drives about once every three years, so if the SSD lasts that long (it's warranty period) I'm fine with that and the increased operating capability it provides has then more than paid for itself.
Was so impressed with it in the laptop, bought a budget version for the travel netbook I use. Same pattern of usage. (granted the chipset there becomes more of the bottleneck).

 

You're welcome

It really depends on what you mean by reliability. If, say, I am buying a netbook for a 12 year old that takes it around with him. A cheap SSD will last longer than a HDD. The HDD might not survive the first drop down the stairs.

However, if you have a system sitting on your desk, not moving - the HDD may very well last longer. Note that "wearing out" is only one of the modes of failure. The support forums are full of posts about SSDs that just one day drop off the radar and can't even be detected by the BIOS. Or, one day all the data is gone. These kinds of things happen because SSDs are still a relatively new technology. There are usage scenarios that the manufacturers haven't been able to test yet, and real world users are running into them.

 

Yes but when a HDD fails, it will cost thousands of dollars to extract that information, if it's even possible to extract it. Remember a mechanical hard drive only has so many kinds of failure and just about every one of them are hardware related. A SSD once it's rewrite operations have been pushed to their limit, they still retain the information that was last saved. Thus you can still read from them, you just can't write or even clear (unless you pass massive voltage through it or pass a powerful magnet). The data is easily extracted and even if professionals get involved shouldn't cost more then a few hundred dollars.

That's why I mentioned SSD have an edge in security. Security covers a lot of aspects from durability (broken mechanical hard drive or SSD)=not a secure method for storing information) to encryption (both are equal with SSD taking an edge if hard ware encryptions are added on the board which can be bypassed on mechanical hard drives via platter extraction), etc.

Overall SSD's are more secure then HDD's for purposes related to permanent storage with little to no writing operations to the SSD.

However I do agree that their life expectancy is largely exaggerated, and to be honest does not warrant the price difference solely based on longevity.

Manufacturing defects will happen on both mechanical hard drives, and SSD there is no avoiding that.

 

some people just want an ultra-thin machine. nothing wrong with that.

 

Crimsoned, you're right. Yes, data can still be extracted. A small problem though, about permanent storage - an HDD can hold data on the platter indefinitely. An SSD, on the other hand, due to electron leakage will lose data over a certain number of years sitting on the shelf.

Oh and guys, please don't turn a thread about SSDs into Apple vs PC