SSD = Failure

[ntrceptr]

I've gathered the following info on SSDs

1. They have a limited number of re-writes per cell (3000-6000 times).
I've found that to keep the drive from failing too soon the firmware of an SSD actually writes to a different location when re-writing a sector to try and lengthen the drives lifetime, keeping one chuck of cells from getting over used. Where it wastes space to track this information I do not know yet. This also means that the fuler a drive is the sooner it will fail as there is less free space to distrubute writes amoung.

2. Enterprise SSDs will only retain information for 3 months if it is not re-written. Others only 12 months.
Question here is if im using this for my OS and that particular file(s) is not updated for whatever reason won't it forget the data after said time has passed and my system will fail to boot.

3. Why on earth would you use this for a Windows OS drive with your swap file on it. It will surely fail much more quickly. I understand the reason for using an SSD for your OS but your swap file should be on a real HDD. Yes this will hurt your performance some but keep the constant writes off the SSD.

Could someone (*wink *wink Toms!) stres test some SSDs until they fail. Be sure the drives are nearly full and write and re-write the free space until it fails.

I was considering getting an SSD until the little details popped up.
Plz clarify to me if I am wrong.

 

[CsG_kieran_2]

1. that takes about 15 years to happen(depending on model). Cells can also be "regenerated" back to 100%
2. Hmm... I don't know about that.
3. SSD are equip with software to reduce the amount of "failed" read and wrights, And that ties into No1.

 

[ntrceptr]

Some links to information

referencing #1
http://www.xbitlabs.com/articles/storage/display/flash-ssd-forecast_4.html
If a drive was near capicity wouldn't this threshold be reach quite quickly?

Referencing #2
http://www.xbitlabs.com/articles/storage/display/flash-ssd-forecast_6.html
I was re-reading this and it did say in the turned off state....so i guess as long as you use it it won't forget anything

Reference to #3
No link - just personal experience with flash memory
I have had Thumb drives fail after 1-2 years....these drives have some spreadsheets on them that get updated every few days. Now they have errors. Different memory type?

New info:
How does one go about refreshing the cells back to 100%
Also is this related to the performance degredation they have over time. I read they actually perform slower the more they are used.

 

[blackhawk1928]

^Uhm buddy...SSD's last much longer than HDD's generally. It does depend on the workload and tpye of HDD and SSD but the general thing is that a SSD in your desktop will outlast an HDD. Your 3000-6000 writes is inaccurate. First off, it depends on the type of SSD based on brand and version (SLC and MLC). SLC is said to have up to 100,000 cycles per cell and MLC is over 10,000. I don't get what you are complaining about, to write even a 1,000 times to each cell of your SSD assuming it were 100GB you would have to write 100TB of data to it. IF you have difficulty realizing what 100TB, thats 100,000GB. So if you wrote 10GB a day to your SSD which I doubt you will unless you constantly install large programs then it would take 30 years to max out a 100GB SSD drive with just a 1,000 writespercell cycle...now imagine the SSD's which have 10,000 cycles like most do.

Simply what I said above, put it this way.

BUY AN SSD, SHOVE IT INTO YOUR COMPUTER, AND FORGET ABOUT IT FOR THE NEXT 10 YEARS.

 

[CsG_kieran_2]

^ +100TB

 

[randomizer]

1. The number of program/erase cycles varies depending on the NAND. SLC is usually rated around 100,000 cycles per block, while 50nm MLC is ~10k and 34nm is ~5k. SLC is being gradually phased out even from the enterprise market though because it is simply not cost-effective. It is cheaper to use more MLC to offset the poorer endurance than use SLC. A drive that is always near capacity but which receives a lot of use within the remaining free space may suffer uneven wear but it won't necessarily die any faster than a more uniformly used drive. It will potentially lose capacity faster though as the overused cells will fail before the rest.

2. The controller will shuffle data around over time, even if you never explicitly do anything with it, and even if the drive is nearly full (it's just harder in this case). It just needs power

3. I run my swap file on my SSD. Swapping is slow enough as it is, I wouldn't want to make it any slower! There's obviously a performance/lifespan trandeoff here. I don't expect that a swap file will devastate your drive though, unless you've got no RAM and you're trying to run photoshop 😉

 

[adampower]

Did you read the toms article with a pickup truck full of memory? It talked about having enough memory to shut down the swap file. They said 12gb is usually enough but 16 is definitely enough. Next rig will have 16

 

[blackhawk1928]

I don't even have a pagefile...with enough ram, i see no reason to have it. And randomizer, so the newer SSD's with 34nm have a lower lifespan than the older ones at 50nm?...why would newer SSD's have lower cycles 🙁?

 

[randomizer]

There is lower lifespan per block, but they have more blocks The controller logic is improving as well and controllers are able to more efficiently write to the drives. Sandforce's latest controllers, for example, attempt to compress data before writing.

 

[ntrceptr]

Thanks to everyone for their input/experiences. I questioned this due to my flash drives failing so easily and from a response Toms E-Mailed me on having on SSD fail on them (they need to investigate the failure still though).

I found the same 100k cycle per cell on wikipedia too....wikipedia is not fact! I prefer the data spec from the manufacturers
WDC states here that they are reducing the cycle (cell endurance) so they can make them cheaper. So yes, some of the newer cells have a lower cycles.
http://www.wdc.com/WDProducts/SSD/whitepapers/en/NAND_Evolution_0812.pdf

And how can anyone say an SSD typically lasts longer than an HDD. They have not been on the market long enough, unless you did some sort of test with an SSD and HDD to see which failed first.

I'm not trying to argue say anyone is completely wrong. I would like to see a stress test.

 

[adampower]

The thing about SSDs is that we know how long a specific nand cell will function (barring any unnatural circumstances). Therefore one can make an accurate estimation of the lifespan of a given ssd. HDDs are a little for difficult to pin down as they have so many moving parts.

 

[randomizer]

I prefer the data spec from the manufacturers

Even manufacturers can only give you an estimate. Because a cell is rated for 10k or 5k doesn't mean it will last for exactly that many cycles. It could be hundreds or even thousands more or less, and it will vary from cell to cell. Some will much much closer to spec than others.