What are the longest life expectancy you have seen in SSD drives?

[SRussell]

I am considering moving my critical data to an SSD array.

If you are looking at 4x Samsung PM1633a, or equivalent drive, in a RAIDz1 array; do you think you could easily get 10 years worth of life from the array? I have 21Tb of total data, adding 3-4Tb per year, with not much write and rewrite.

I know I will never reach 20% of the stated DWPD to TBW. The drive claims 2,000,000 MTBF. These would be used drives with 10-150DWPD.

 

[i386]

Does this count?


I think with enterprise ssds you will run out of space before you kill the devices.

 

[T_Minus]

Some thoughts:

• run RaidZ2
  
• try to find MLC drives, the problem is capacity needed and overall choices of "new old stock" may be limited
• make sure you have enough cooling, all the time if you want long-life the enterprise drives run HOT
• I personally wouldn't run 4 or 6 7TB drives for example for a long-term\life, I would run more smaller capacity (not 24 but maybe 12-16)

 

[acquacow]

I personally have "ssds" that I've been using for over 10 years, but they are enterprise Fusion-io memory modules...they will probably outlive me. I've had them running since 2010.

The least reliable ssd I have is a 1TB Samsung 840evo that I've had in service since 2014 and it's still chugging along just fine.


This is the nice thing about flash, it either dies upfront from infant mortality, or it lives a long life.

 

[Marsh]

21TB data , 10 years x 4TB year ,total = 61 TB data

I have few old Intel 320 series 600GB SSD that had 81TB data written, it was showing 95% life left when I retired the drives because of slower SATA2 interface.

 

[EffrafaxOfWug]

They're not under heavy use, but I've still got plenty of ~10yr old SSDs in active service (including a couple of OCZ Agility 120GB as boot drives somewhere). The biggest cause I've seen of SSD failures has been either premature controller death or firmware bugs, and I've been lucky enough not to have any failures at all in the last 5 years (I've not really been buying any consumer-level drives for myself though). As aquacow says, the bathtub curve for most SSDs is an exceedingly long middle section and most SSD models we don't know what the right-side curve even looks like yet. I suspect the SATA interface will be made obsolete before the majority of my SSDs are.

If I had the money for an all-flash array currently, I would use the Micron 5200/5300 semi-enterprise TLC SATA drives; low power and temperature (and speed) compared to SAS drives but available up to 7TB, relatively cheap, good enough endurance rating and a good reliability provenance.

 

[ajs]

Does this count?
View attachment 13019

I think with enterprise ssds you will run out of space before you kill the devices.

Only because that drive is so ridiculously big you can't physically write it fast enough to put enough PE's on it to wear it out within its warranty timeframe.

 

[Octopuss]

There was a two years or so long endurance test of some SSDs a few years ago. Crazy results.

 

[BlueFox]

I still have a 64GB SLC SSD ($1000 back then) from my Lenovo X200 that I got in 2008. It reports 91% health 12 years later.

 

[zack$]

Is cold storage a factor for anyone?? I think this is one caveat that should be mentioned with respect to ssds and data retention (which I admit is not the same as "life expectancy" but related nonetheless).

 

[BlueFox]

Is cold storage a factor for anyone?? I think this is one caveat that should be mentioned with respect to ssds and data retention (which I admit is not the same as "life expectancy" but related nonetheless).

JEDEC spec for consumer drives has fairly decent retention. It's a function of the active drive temperature (when data is written) and the storage temperature. Every 5C drop in power off temperature approximately doubles the retention and every 5C increase in active temperature gives you approximately 50% longer retention. So, ideally, you want high power on temperature and low storage temperature.

For example, at 45C active and 25C storage, you can expect at least 3 years retention. 25C is fair bit above room temperature, so, I would expect 5+ years at 20C.

 

[acquacow]

JEDEC spec for consumer drives has fairly decent retention. It's a function of the active drive temperature (when data is written) and the storage temperature. Every 5C drop in power off temperature approximately doubles the retention and every 5C increase in active temperature gives you approximately 50% longer retention. So, ideally, you want high power on temperature and low storage temperature.

For example, at 45C active and 25C storage, you can expect at least 3 years retention. 25C is fair bit above room temperature, so, I would expect 5+ years at 20C.

That's going to depend heavily on flash design/lithography.

For instance, 10nm floating gate flash only holds about 75 electrons per cell with an electron leakage of around 2-3 electrons per month... Not sure I'd shelve that longer than 3-6 months.

The newer 3D nand stuff has much leakage, but still...