Durability - MTBF value correspondance to TBW

Internal SSD All Other Internal SSD
1

I bought a Extreme 240 GB with has according to SanDisk 2 million hours MTBF (that would be more than 200 years!?). 

Usually on SSD the durability / maximum life time is measured in TBW (TerraBytes Written) - a nice and easy to understand value. Additionally for customers it is easy to read out the current value via SMART - acording to my observations the value “Lifetime writes to host” shows in the RAW value the number of Gigybytes that has been written to the SSD in it’s life time.

Therefore I would like to know if and how the MTBF value can be translated into a more useful TBW value.

2

Just my guess but i think there is no relation between MTBF and TBW. I think MTBF refer to any kind of failure to work properly and that could happen regardless if you are reading, writing or idle. That in turn means tere is no exact relationship between MTBF and the amount of data writen. MTBF seems to be a measure of reliability relevant in the enterprise business where you have many drives. For example 2m hours MTBF would mean something like if you had 230 drives in a server, you can expect one to fail each year.

Of course, it’s my theory and i could be wrong as well. Anyway i hope it can give you a clue. On the other hand, it would be great if someone at sandisk could give us further detail regarding those specifications…

3

@memrob wrote:

I bought a Extreme 240 GB with has according to SanDisk 2 million hours MTBF (that would be more than 200 years!?). 

 

Usually on SSD the durability / maximum life time is measured in TBW (TerraBytes Written) - a nice and easy to understand value. Additionally for customers it is easy to read out the current value via SMART - acording to my observations the value “Lifetime writes to host” shows in the RAW value the number of Gigybytes that has been written to the SSD in it’s life time.

 

Therefore I would like to know if and how the MTBF value can be translated into a more useful TBW value.

Hi memrob,

They are basically saying that they have a 2Million hour MTBF based upon Telcordia Stress Part testing… which is a rating of electronic component failures if kept within the enviromental specifications of the unit in question. 

Here is some info on the Telcordia MTBF rating…  http://www.reliabilityeducation.com/intro_bellcore.html

It looks like the Sandisk X100 MLC drive and the enterprise MLC SAS drives do have TBW ratings.   

I am currently assuming that it is not in Sandisk’s best interest to publish this information.

I have known about the difference between SLC / MLC ever since the industry started selling “consumer level” SSD’s…  My first SSD was a Samsung SLC SSD (Pre TRIM) and it was will over $1000…

Take a look at this review of the Intel 520 SSD (Which I consider to be a comparable drive to the Extreme)… pay attention to the pros and cons section… http://www.storagereview.com/intel_ssd_520_enterprise_review

The con’s state that the 520 MLC can’t compete with the endurance levels of eMLC… (or SLC of course)

I thought the whole idea of eMLC was that a drive of that classification was just a “Short Stroked / Overprovisioned” version of an MLC drive that was not overprovisioned by the factory.  The overprovisioned area is reallocated to account for failed cells, and when there is much more “Spare MLC” available, you could actually meet or exceed the reliability ratings of an SLC drive of the same amount of storage.

So … a 100GB eMLC drive would have 28GB of Spare Area… the 120GB Sandisk Extreme has 8GB Spare,  a 128GB MLC drive has little to no spare area…  but both have 128GB of comparable MLC NAND.     Also, when eMLC was introduced they were some of the first drives to have SuperCAP’s on them to prevent data failure from power loss.  Now most of the new SSD offerings have Super CAPs for power loss failure.

If you take a look at the OEM Sandisk X100 SSD, they are quoting 80TBW under 128GB/day workload, and the same 2,000,000 hour MTBF as the extreme.   (Does anyone know how close the X100 is to the Extreme, what controller, etc?)

I really would like to know whether or not the Extreme will allow you to reset the controller to a 20% overprovisioned mode.  I may have to dig up someof the low level HDD utils and do some experments soon, see what can be done… I would love to get an extra 20GB of spare in the Extreme and throw it into a high workload application.  My gut tells me it will hold up well… unfortunately the official specs for the drive are keeping their vow of silence.

My current opinion is that the Sandisk Extreme SSD line represents an amazing value as far as I am concerned.  If your application requires a drive with a manufacturer published TBW rating, then you may want to look at drives that are publishing this specification…  

If there are any guys out there who have poked around inside the 2281, or done some low level settings on this drive, or other SSD controllers, or have any comments… please chime in…  Just don’t say anything that would trigger a shortage on the Extreme…  :-)

Personally…  I love a bargain…  and I love this SSD… 

Rich

4

I’ve been testing one of the sandisk extreme G25s for how many writes you can pump onto it before it fails.

http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm&p=5121415&viewfull=1#post5121415

For the purpose of this discussion, I’ll keep it to how many writes you can do before the drive reaches the end of its rated write life (MWI = 10, given no bad blocks)

To answer your question more fully though, how many writes you can send to the drive before it’s defined lifetime is over is a question that depends on a few variables:

a) How busy the drive is: If the drive is busy, then user host writes will be the majority if writes incurred, and that will be the determinator for lifetime. If the drive is very quiet, but powered up for many years, the internal garbage collection and refresh circuitry will be what contributes more of the writes.

b) How compressable your workload is: The more compressible your workload is, the more writes you can send to the drive before lifetime is over.

c) The write pattern of your workload: Desktop loads typically generate lower write amplification, Server loads typically create high write amplifcation.

Now, the drive I was testing is using somewhat compressable data, with a write pattern of a desktop load, and I was able to write approximately 435 TB (Binary Terabytes) of data to the drive before the drive reached specification End of Life. I would say most desktops would get at least 350TB written before end of specification life. You can double that figure for 240GB drives and probably triple that for 480TB drives (480GB drives have a larger logical block size, so I will speculate cautiously for that drive)

However, depending on your particular workload, you would get anywhere between 250TB and 500TB written before specification end of life. Not only that, but seeing how the NAND (24nm Toggle) in these drives has done in other drives, you may get twice the specification life before the drive actually fails.

You can monitor roughly how many GB of NAND writes your drive has done by looking at the raw value for smart attribute E9. My Sandisk Extreme 120gig reached end of life when that number reached around 330,000, and I would assume other 120gig Sandisk Extreme drives reach end of specification life at the same point, regardless of user load or work pattern.

5

@rchgrav wrote:

To answer some of your questions:

eMLC is a combination of selecting the best MLC NAND dies, applying very strong ECC correction, relaxing retention constraints and good bad block management … that allows Intel to create drives with much higher write endurance then standard NAND, while not significantly changing normal MLC NAND production.

If you want to overprovision a desktop drive, you can do it realitively easy by secure wiping the drive, then only partitioning how ever much space you want to use. The remaining unused LBAs will never get used, which means the drive will be able to use the left over NAND to perform wear leveling and write optimisation, reducing write performance degredation under some loads. However, desktop loads under windows 7 generally don’t fall into the catagory of requiring additional overprovisioned space for performance.

Also, I agree that the sandisk extreme has been amazing value for a high performance drive.

6

Thanks a lot for sharing that useful information. Now I know that my Sandisk Extreme SSD will last long enough even if the chips are 24nm instead of 32nm