The SSD Endurance Experiment: They're all dead

So it looks like even consumer SSD should last at least 300TB of writes

6 SSD tested, 2 lasted over 2PB of writes before dying.

http://techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead

John.

 

Need to remember that just because you can fill an SSD with 2PB of writes, doesn't mean it will last that long in a real world workflow.

As an example; you don't train for a marathon by swimming (exclusively). You need to run.

The workload an SSD performs is made simpler by the algorithm the firmware has - and as long as it doesn't need to switch gears to a different (for the SSD) workload - the firmware becomes more efficient for that specific workload (up to it's maximum).

What those tests did was confirm that the nand is more robust than was initially thought. But that is a far cry from the nand performing as expected, the power cycling (and hard power offs) and the many different workloads (not workflows...) that the SSD as a complete unit must endure in a typical lifetime.

I have killed 60/80/128/256GB SSD's in a week (performance-wise), never to be resurrected. I also have SSD's that with more than 50% OP'ing have been working for years (as Scratch Disks for PS and other programs I use daily that can use a dedicated 'temp' drive).

This kind of 'record' is deceiving because an SSD's lifespan depends on the workload it is designed and used for. If they can just use the same drives with real world workflows/uses that simply hammers the drive non-stop (unlike what an owner can do by themselves...) with no speed up of the script - real time is important here (for TRIM, GC, voltage swings and temperature fluctuations in both the nand and the controller chip + algorithm that takes (or not) all this into consideration.

To me, this is equivalent to the capacity to throw a man to the moon. But, forgetting about keeping him alive or bringing him back home...

Whether that 'moon' is 300GB or 2PB away...

Take care.

 

I already knew that SSD life was a moot point for the average consumer (& me!), but it was interesting & surprising to see JUST how much extra life can be in our drives. I know though that I'd be swapping out my SSD if it ever got to 0% Drive Remaining Life on the SMART Attributes. My 840 Evo that I have as a system drive still has 100% Drive Remaining Life after 1 year 3 months usage and 7.5TB written, I'm quite confident that they'll be obsolete well before running out of life!

 

How much data can be written to an HDD before it dies?

When you say die, do you mean it stopped reading, stopped writing, or simple stopped functioning and was no longer recognized by any system?

Sent from my XT1049 using Tapatalk

 

Magnetic hard drives wear differently than SSDs. While SSDs have a limited number of times each NAND block can be erased and re-written before data becomes corrupted, hard drives have no write limit. But since hard drives are susceptible to friction wear, temperature, altitude, and physical shock (unlike SSDs), they generally fail faster than SSDs.

The way SSDs are supposed to "die" is they can no longer be written to, but they can still be read. The way they tend to actually "die" is they become unrecognizable to the system.

 

Thank you djembe!

Sent from my XT1049 using Tapatalk

 

Great answer djembe.

Would also add that a HDD will generally give an indication days, weeks or months before it fails; SSD's? About a nanosecond.

And for me, a storage subsystem failure is not only when it stops reading, writing or responding at all - I also include when the performance/productivity of the system has fallen below what I could achieve with a 'mere' HDD...

That fact above is what kept me from buying and keeping/using the first couple of years worth of SSD's I tested (starting circa 2009...).

Failure is not simply an on/off 'condition'. The shades of grey are just as important too.

That is why Samsung SSD's with their laggy performance (yeah; at more than 'light' loads) leave much to be desired...

 

Performance is one thing, endurance is another. One thing I never really cared for was disabling or removing the hibernate.sys file as I would read it would caused unnecessary wear and tear. Same goes for defragmenting an SSD. I use hibernate all the time, I find it very useful so I wasn't at all worried about minor wear and tear. This experiment goes to show hibernate won't make a dent to its endurance. I'll keep my hibernate.sys file why thank you. Lastly, wasn't it you tiller that showed defragmenting an SSD actually showed some tangible benefits?

Sent from my XT1049 using Tapatalk

 

Yeah; guilty.

Although I don't know if I actually 'showed' anything? Just reporting that defragging an SSD is just as important as defragging an HDD if high performance and responsiveness is important.

See:
http://forum.notebookreview.com/thr...ion-are-there-now.784638/page-2#post-10173580

 

So the winner here was clearly the Samsung 850

 

Defragging an SSD and any apparent noticeable improvement is not tangible nor long lasting. SSD's like to move data frequently for wear leveling and GC will kick in too which will move stuff around. SSD's know where they want stuff. All it really does is organize the file allocation table better which may result in slightly improved access times, but where the data is on the SSD only the SSD knows. If it were that tangible and significant improvement then SSD makers would offer such a tool because it would make their product seem faster than it actually is.