Stability/Reliability testing for SSDs?

[Boomhowler]

Hello!

With the current top 5-6 SATA 6GBps SSDs performing more or less equally for the average user, the statistic I am getting more and more interested in is stability and reliability. I really enjoy the stress-tests Anand have shown in the review for the Samsung 830 for example, these are very important for me and what I want out of a drive that's supposed to contain the OS and software of my workstation.

Anand (and everyone else), have you been thinking about how to conduct such "over time reliability"-tests to see for example how well the "MTBF"-numbers compare to what's advertised by the individual companies?

If not, does anyone know if such information (return statistics etc) is available anywhere?

 

[R4in]

Im not sure this kind of information would be widely available.

For one, when groups such as AnandTech, TomsHardware, etc get things to test, they dont normally get to keep them for more than a couple of weeks. You are looking at tests that may require upwards of 6 months to a year of constant testing.

Also, if you DO want info like this, users are going to be your best bet. They may be as wide-known as above groups, but they could potentially be just as reliable since they get to keep the drives for quite some time (until upgrade or it craps out).

Best thing to do is just see what everyone else is buying and recommending and go with it. Really cant go wrong.

 

[IGemini]

Also, if you DO want info like this, users are going to be your best bet. They may be as wide-known as above groups, but they could potentially be just as reliable since they get to keep the drives for quite some time (until upgrade or it craps out).

This. The reason why reviewers at AT and other places flood the SSD with I/O to test TRIM is to emulate drive wear. It's the same basic principle that companies use to get MTBF values since neither have the luxury of actual time.

It's really hard to get any number of people who are objective about longitudinal data. First impressions or cross-sectional are hard enough to take seriously with some user reviews: "I bought this SandForce SSD and it corrupted after the computer hibernated, SSDs suck."

 

[Emulex]

"I bought this SandForce SSD and it corrupted after the computer hibernated, SANDFORCE CONTROLLERS suck."

FTFY

I have deployed 50+ intel mostly x25-v and had 1 failure in the last 2 year. it was a sandfroce lololol. no j/k it was 1 x25-m and 1 sandforce. they never bothered to patch the sandforce so what good is it? like a smartphone/tablet stuck on android < 2.3.4 - wasted

 

[IGemini]

^^I was intentionally generalizing to mock such amateur reviewers. Notice I do use a SandForce SSD.

 

[Boomhowler]

Yes, I understand why it is difficult to produce such tests and that such information is sensitive to the companies but it WOULD be interesting to see I saw some french site which had some numbers from last year with Intel leading the return rates with like 0.6% and all the others were situated above 2% but Samsung was not included in the mix which I see as the main contender to Intel on this point.

The thing I wanted to avoid was just that anecdotal evidence with "users reporting". It's mostly the people who have problems with something (or are fanboys ) who write lots and lots on the internet about such things which makes the statistics skew in the best case and simply wrong normally...

I guess that warranty time and the reported MTBF is the best way to choose a stable drive for now and then when you buy a drive you run a couple of stress tests immediately to see if all blocks are ok etc.

 

[Sunburn74]

I think this is a very interesting test. It doesn't really address the issue of stability but certainly does for the issue of reliability

http://www.xtremesystems.org/forums/showthread.php?271063-SSD-Write-Endurance-25nm-Vs-34nm/page9

Essentially this guy developed a little program that would write to his kingston 40gb intel ssd constantly at a clip of about 2TB of random writes a day non stop. He then followed the drive to see how it does performance wise and if it'd live up to intels claim of 20gb a day for 5 years (or 36tb lifespan). At page 9 he's at 25.72 TB of writes on a 40gb drive and its still plugging along (media indicator had barely dropped. At around page 12, he's at 42.95 TB and still plugging along (wear indicator in the low 70s/ if wear indicator hits zero, the drive should be at its end). I have yet to finish the entire thread as its 90 pages long but the point is clear: wow.

I think its amazing that he got 42TB out of a 40gb drive without breaking a sweat. I have 2 80gb intel drives in raid and have had them for about 2 years now. My total writes is 4TB per drive in 2 years.

In addition after following the thread, it seems like the realistic lifetime of an intel 40gb drive (or any ssd with a write amplification around 1) will be close to 250TB roughly. One guy started the same experiement with a 60gb vertex drive and was at 118tb around page 18.

Edit: Wow. At page 78, his little Kingston 40GB (X25-V) had managed 362.4TB of RANDOM writes without failing or losing speed significantly (roughly 10-20percent drop in seqential write speed). I am absolutely blown away. 20gb a day for 5 years is a huge huge underestimation. I now feel absolutely free to totally abuse the hell out of my intel drives.

 

[RU482]

I'm running the test Sunburn74 linked to on a Micron and a Sandisk SSD. Both are still purring along, sitting at just under 30TB written.

I'm letting them run as long as it takes.

I'll probably add a few more drives in the near future, if I can get another rig going.

 

[Boomhowler]

Sunburn74, that's a very cool test! Unfortunately they are both very time consuming and it will (perhaps) ruin the drive, and it only tests one drive. It would be cool if intel, ocz etc. got together and tested a couple of hundred drives each in a similar fashion so we could get some significant statistics out of it. Not that they would ever release those numbers live.

Hopefully will everyone that have run this test gather the info in some sheet where you could see which drives that perform best over time etc.

On the other hand, the number these tests produce are fantastic and as you say, I will never be worried that my SSDs will start to degrade due to lots of writes