An interesting read about SSD endurance

Hello everyone,

I found this interesting article about SSD endurance experiment on the Tech Report.

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

Good read, have a look.

 

The endurance of flash has not been an issue for me for a few years now (since ~2011 when I started defragging my SSD's). But still surprising that they lasted this long, of course.

What is most telling to me is the ASU average write speeds;

See:
http://techreport.com/review/27909/the-ssd-endurance-experiment-theyre-all-dead/3

(Last graph on the link above).


Maybe now people can see for themselves how miserable Samsung's SSD's rate for 'performance' vs. better options. And why I call them 'laggy' since the first time I've used one almost 5 years ago...

This is simply not SSD levels of performance. And, depending on what HDD you compare to, not even HDD levels either. (And forget the TLC example, it's just sad - at least the 'Pro' tries to get higher than 125MB/s once in a while).

The Corsair with it's Link A Media Devices (LAMD) controller is a real surprise here. Still, only half of the 'up to' write speeds SSD's promise, but with nothing bigger offered than 256GB (just locally?), it was never in consideration for my needs anyway.

I never warmed up to the Intel 335 Series in my own systems (the 520 Series were superior); but it gives great consistency and almost double the Samsung's 840 Pro's efforts, right to the end. Even with the much hated SF based controller.


I hope this starts to put an end to the people that say that any SSD will do, vs. a HDD. That is simply not true.

When you're paying at SSD levels (even at 'sale' price levels) vs. a HDD and you get HDD or worse performance, it would have been better to stay with an HDD or save a month more to get a real SSD.


I agree 100% with the article though about the first sign of trouble is the time to retire an SSD. No warning, no way to recover any data (without a $1-4K data recovery charge) and nothing but the memories of when the SSD would boot your system up and life was good.


Thanks for the link and the thread.

 

Looking at my modest 128GB Crucial m4, I've managed to consume 196 P/E cycles over the last 2.4 years, so just under 0.25 P/E cycles a day. At this rate I'll have another 30 years before it wears out. Sweet!

 

I wonder how operating temp. affects those numbers? They probably run those tests in controlled environment, the msata drive inside my laptop runs at about 50-55c and it's still winter here, summer comes, the drive will be hitting 60-65c for sure, may need to put some fan inside to cool it off? I wish there was some way to put some heatsink at least?

 

That was an interesting read indeed.

I just checked in my Linux system with the iostat command that I have written 233722224 sectors of 512 bytes in the last
12 1/2 days. This roughly translates to a total of 9GB written per day, and a total of 3.25TB per year.

If we assume a minimum endurance of 200TB before even a reallocated sector appears (as suggested in the article), an SSD
would last about 60 years.

Seeing that most of the SSDs reach 500TB without significant problems, I could probably squeeze 150 years out of my SSD.

Impressive.

 

I suspect some other failure mode will present itself with SSDs before anyone gets literally to "60 years" with minimal writes. Solder joints. Capacitor failure. Etc.

 

Well, yes, clearly. I just think that "running out of writes" is not a realistic concern for the majority of uses cases. I mean
even if someone wrote 100GB per day, like every single day, which seems very extreme to me, she would get about 15 years
of life.

I wonder if an HDD has been tested under this endurance experiment. Is an HDD mechanically able to endure 1PB of writes
or will it collapse long before that?

 

I doubt a standard desktop drive would endure that much. Since I've been running a home server of some sorts for the last ten years, I can attest to that fact. Even my drives didn't get a lot of use, just that they were on all the time and powered up and down regularly for scans, and backups, and defrags, etc. Desktop drives fail 50% of the time within 3 years, and 100% within 5. Those are a mixture of WD, Seagate, HGST, Toshiba. Server and NAS rated drives (WD keep on chugging along though. WD red drives still managing fine after about 3 years, while in that same time have had 2 of my 6 Seagate desktop drives fail. Intel SSD that is the OS drive, keeps on chugging along as well.

And SSD's are known to have failed controllers long before any NAND wear issues. But it's good to see that NAND wear can finally come off everyone's worry list. Just go with MLC drives still.

Too bad there wasn't a Crucial SSD in that test though. Would be good to see their performance in the end. As well as some TLC drives.