SSD reliability in the real world: Google's experience

[Brink]

Using data from millions of drive days in Google datacenters, a new paper offers production lifecycle data on SSD reliability. Surprise! SSDs fail differently than disks - and in a dangerous way. Here's what you need to know.

SSDs are a new phenomenon in the datacenter. We have theories about how they should perform, but until now, little data. That's just changed.

The FAST 2016 paper Flash Reliability in Production: The Expected and the Unexpected, (the paper is not available online until Friday) by Professor Bianca Schroeder of the University of Toronto, and Raghav Lagisetty and Arif Merchant of Google, covers:

• Millions of drive days over 6 years
• 10 different drive models
• 3 different flash types: MLC, eMLC and SLC
• Enterprise and consumer drives

Key conclusions

• Ignore Uncorrectable Bit Error Rate (UBER) specs. A meaningless number.
• Good news: Raw Bit Error Rate (RBER) increases slower than expected from wearout and is not correlated with UBER or other failures.
• High-end SLC drives are no more reliable that MLC drives.
• Bad news: SSDs fail at a lower rate than disks, but UBER rate is higher (see below for what this means).
• SSD age, not usage, affects reliability.
• Bad blocks in new SSDs are common, and drives with a large number of bad blocks are much more likely to lose hundreds of other blocks, most likely due to die or chip failure.
• 30-80 percent of SSDs develop at least one bad block and 2-7 percent develop at least one bad chip in the first four years of deployment.

Read more: SSD reliability in the real world: Google's experience | ZDNet

 

[whs]

Interesting article although I am not sure what to draw from the conclusions.

 

[ThrashZone]

I guess I got a little taste of the bad block syndrome that went on to kill it's self
Another one on the way on sure if it's new or refurbished the invoice is pricing a new @ 134.22 sure hope Crucial expects payment

I take it from the article backup..... more news at 11:00..

 

[Layback Bear]

Well it is interesting information to me.
But like whs I don't know what I'm going to do with the information.
I'm quite sure my ssd's will never be used as hard as a Google data centers ssd.

I would like to know what ssd's were tested and how many wrights and reads were done and how old the ssd's were before things starting going south. I presume that these ssd's were in a temperature controlled rack somewhere in the world.
I wonder if storage capacity and free space showed any difference on wear and tear.

Na it really doesn't matter I'm going to keep buying Intel because they work for me.

 

[whs]

Jack, we are the two clueless in this thread, LOL.

 

[Layback Bear]

Yes for sure I belong to the clueless, but I'm curious.
The article gives just enough information to tickle our curiosity.

 

[profdlp]

Key conclusions

• Ignore Uncorrectable Bit Error Rate (UBER) specs. A meaningless number.

No problem, I will.

• Good news: Raw Bit Error Rate (RBER) increases slower than expected from wearout and is not correlated with UBER or other failures.

Nice.

• High-end SLC drives are no more reliable that MLC drives.

This is the most useful bit of info here, if you ask me.

• Bad news: SSDs fail at a lower rate than disks, but UBER rate is higher (see below for what this means).

If the UBER rate is to be ignored, where is the bad news here?

• SSD age, not usage, affects reliability.

I'd count this as a plus, since age is easier to keep track of than usage. A negative would be that you might be wasting your money by putting an SSD in a computer with light usage you plan to keep for years.

• Bad blocks in new SSDs are common, and drives with a large number of bad blocks are much more likely to lose hundreds of other blocks, most likely due to die or chip failure.

The same is true for mechanical drives. Once a mechanical drive develops bad sectors it's time to schedule its replacement - and soon.

• 30-80 percent of SSDs develop at least one bad block and 2-7 percent develop at least one bad chip in the first four years of deployment.

Sorry, but any statistic mentioning a range of 30-80 percent tells me they really have no idea what the actual number is. 30-80 percent of girls might say "yes" if I ask them out on a date. Should I make dinner reservations?

 

[Borg 386]

I know SSD's are a lot faster, but from my perspective, I don't see a reason to get one because of the high prices & they last about as long as a regular HDD. When the prices come down & they start outlasting the spinners, then I would opt to get one. In the meantime, I'll just wait the few seconds longer it takes for my OS to boot.

 

[maxseven]

SSDs fail differently than disks - and in a dangerous way.

This is all I need to know about SSDs (and I knew this already as I've had defective SSDs in the past).

The rest IMO is meaningless unless one is doing something differently like creating a RAID or JBOD using SSDs for some critical purpose. In which case I'm inclined to use Intel just cuz I like 'em.



But for individual day-to-day use, whether personal or work-related, just buy whatever's appropriate at the best price and then BACK THE THING UP REGULARLY! which is always best practice, but AFAIK is downright critical for an SSD which can disappear without a trace at a moment's notice (failure).

 

[Lee]

I Read this stuff with a grain of salt. . .reason being is I have been running an Intel M 80 as my main OS drive since October 2009 without any problems whatsoever. I do keep another SSD cloned just in case. . . As to the cost, that is becoming a none issue. A 1terabit is cheaper then the M 80 was six/seven years ago, thus to me that is just a lame excuse to not spend the money for one.

 

[Barman58]

I'm not certain when the SSD became generally available but I know that it's generally fairly recent when compared to "Spinners" - as with any new technology reliability increases as products mature, new tricks are found and employed Etc

Taking the above into account I doubt that the SSD industry will be failing in the near future

Also this is from the desk of Ed Bott, so is immediately suspect considering his record of inaccurate scare mongering to earn clicks for his blog

 

[ThrashZone]

Ed wouldn't do that

 

[whs]

I am with you Lee. I have nearly a dozen SSDs with the 2 oldest from 2008. I had 1 little problem with a Mushkin that was immediately replaced. Else they work like clock work.

 

[LMiller7]

I'm not certain when the SSD became generally available but I know that it's generally fairly recent when compared to "Spinners" - as with any new technology reliability increases as products mature, new tricks are found and employed Etc

Taking the above into account I doubt that the SSD industry will be failing in the near future

Also this is from the desk of Ed Bott, so is immediately suspect considering his record of inaccurate scare mongering to earn clicks for his blog

There is little doubt that some form of solid state technology, probably different than is now being used, will eventually completely replace conventional drives. But we still have a way to go before that day arrives. As market share and manufacturing techniques improve cost will go down and reliability, storage capacity, and performance will improve. Just as happened with conventional drives.

 

[Layback Bear]

I had one Intel ssd die but I'm not sure if it was the ssd or the connector from the power supply harness. They melted together and let the smoke out.

Really didn't matter. I replaced the harness with a extra one I have and Intel RMA the ssd in 1 week. During my down time I just popped in another ssd and went on with my happy computing. Everything in a computer can have problems. It's how well the company helps you with your problem. I like the no fuss get the problem taken care of method. My dealer and Intel worked great together and things were done asap.

 

[ThrashZone]

Until recently I was thinking ssd's are biotchen
Read about the garbage collection and I'm still kind of :huh: stage
My SSD used to be so much faster... What happened? - Crucial Community

On a desktop PC, simply disconnect the SATA cable from your SSD and only leave the power cable connected. After switching your PC on, the SSD will be in an idle state but still have power so Garbage Collection can function. On a laptop, power on with the SSD installed and enter your system BIOS (please refer to your system manufacturer’s documentation on how to access the BIOS.) Leave the laptop in the BIOS menu for the 6-8 hours.

 

[whs]

The old BIOS trick. You just have to make sure that there is no activity on the SSD. When in BIOS state you need not disconnect any cables. Then the garbage collection can do it's job. But 6 to 8 hours seems exaggerated.

 

[lehnerus2000]

Also this is from the desk of Ed Bott, so is immediately suspect considering his record of inaccurate scare mongering to earn clicks for his blog

The author is Robin Harris.

Perhaps you meant the ZDNet site (Ed Bott hangs out there).

The bottom line is, make regular backups of the entire SSD.

I've found that I cannot successfully restore individual OS partitions on SSDs.
When I have tried to restore an individual OS partition, that OS won't boot (Linux and Windows).

I'm hoping ReRAM will soon eliminate Flash-based SSDs.

 

[ThrashZone]

That is what they gave me after I posted the I/O error in windows disk management and disk part and mini tool just said Bad disk
I also gave them the gpart error messages too.

 

[Lady Fitzgerald]

Jack, we are the two clueless in this thread, LOL.

If you two are clueless, I'm beyond hope.

All seriousness aside, one thing that caught my attention was the comment about uber failure. Big deal, it's well known that, when SSDs fail, they usually completely fail with no hope of recovering data like you sometimes can do with HDDs. So what? No one should ever count on recovering data from a failed HDD or SSD; at best, it's a crap shoot that can cost a small fortune. That's why it's imperative to always backup data.