Understanding the effects of power failure on flash-based SSDs

[Gary Seven]

Just thought I'd post this link that I found on Usenet (anyone remember that?). I'd be interested in your thoughts on this.

http://www.eetimes.com/design/memor...-effects-of-power-failure-on-flash-based-SSDs

G7

 

[Railgun]

All my drives, both OS and storage sit on an Areca card. I do not have a BBU at the moment as I am ok with that risk. Storage is on HDD, OS on SSD. A BBU in this case really wouldn't make that much difference based on the above.

That said, what I would like to see, is a BBU for each drive, in particular for a desktop solution. Given they're 2.5" now, there's plenty of room available for adding a thin, 12v battery stacked on top of the thing. It should be possible to package them neatly, and relatively cheaply. Unfortunately that doesn't address the laptop market so much, but given power loss is a much lesser risk, I think that may be a moot point.

That said, a proper controller with a BBU combined with the above isn't unfeasible and realistically should be used as a last resort if your UPS fails, or for a safeguard for those that don't run a UPS.

 

[ElFenix]

SSDs have capacitors in them to provide enough power for everything to get done when the drive detects power loss.

 

[Railgun]

Apparently not. Capacitors are not the answer anyway.

 

[kbp]

UPS back-up power brick. Saved my rear many times.

 

[npaladin-2000]

Maybe they're designed to fit laptops because laptops have built-in batteries...?

 

[Railgun]

Well, they're not going to make it bigger just for the sake of making it bigger. They're also not going to design it to toss out the standard format for how ever many millions of laptops that exist.

 

[Emulex]

server SLC drives have HUGE capactiors - way way bigger than the intel 320. I mean HUGE. i suspect they are designed to handle a clean shutdown proper.

Why they are 14MM instead of 7mm - the caps (look at a FBWC raid controller) are the same size per drive. i'd say 7mm height is just the caps the rest of the board is like 3-4mm

 

[KingFatty]

In the study, I don't see any mention of how they disabled internal capacitors. They spoke as if no such internal power existing in the SSDs, and they used some switch to just cut the power supply?

So when people say SSDs have some internal capacitor/battery, maybe they are referring to the way the computer's power supply has capacitors that can do this brownout over time to enable the SSD to survive better than just chopping the power supply abruptly?

It's pretty scary how the study documented retroactive corruption of data that was already "safely" written, doing something destructive that would typically not happen in a spinning disk drive. Also, they showed how data could eventually corrupt itself even though it seemed safe, if there was an affect related to power loss associated with that partially safe data. Pretty scary stuff.

 

[bradley]

Not only are capacitors being used currently as the answer to flash power failures, but a single super capacitor could probably power an SSD for days-weeks.

 

[taltamir]

Apparently not. Capacitors are not the answer anyway.

According to the article they are.

Based on the experimental results we obtained from this project, we can suggest some methods to mitigate the effects of power loss. First, since incomplete program operations may corrupt existing data and the bit error rate does not decrease monotonically as the operation time increases, the SSD should be equipped with backup batteries and capacitors that guarantee the program operation will complete when power failure occurs.

Capacitors are recommended in every SSD to solve this problem.

In the study, I don't see any mention of how they disabled internal capacitors. They spoke as if no such internal power existing in the SSDs, and they used some switch to just cut the power supply?

Apparently they baked them too to simulate age...
BTW, this is a school project.

About the authors:
Hung-Wei Tseng and Laura Grupp are graduate students and Steven Swanson is assistant professor at the University of California, San Diego.

 

[darkfoon]

I got the impression from the article that they weren't testing complete SSDs, but rather they were testing specific chips to see how they handled power failure. If you think about it, in order to control power failure so precisely, it makes much more sense that they have a single controller and a flash memory device, rather than a complete SSD with all its complexities that complicate control of power.
Since many SSDs include capacitors, those SSDs are probably less prone to the errors they found in their research.