Micron's "Power loss protection for data-at-rest"?

[unwind-protect]

Power loss protection for data-at-rest... what exact do they do there?

I have an Intel DC drive that has a SMART attribute counting the PLP capacitor tests. I have a Micron 1100 drive here which claims to have it this Micron thing and SMART has no such attribute.

Is that anything useful at all?

 

[CyklonDX]

micron uses ceramic capacitors on their disks.

Those capacitors typically provide from 1ms to 50ms of uptime designed to last enough time to flush the data from on-disk cache/ram to nand modules.

SSD (10-50ms)

nvme (1-6ms)


is it useful? Yes. I'd say its more important in desktop/consumer grade world, than in enterprise. (while being typical enterprise feature)
Many people have corrupted their files/filesystem by unexpected system halts / overclocking / loss of power etc.. in consumer grade hardware, than in enterprise.

 

[mrpasc]

The capacitors are used to offer PLP for „data at flight“, implemented for most enterprise / datacenter ssd‘s. This implementation makes sure that data sitting in the (RAM)Cache becomes flushed to the NAND in case of power failure.
This is often referred as „end to end“ PLP.
This is implemented in your Intel ssd, thus you can see the capacitors value within SMART values. The Micron 1100 does NOT have this kind of PLP implemented, thus no data for capacitors in SMART.

For „data at rest“ PLP is more marketing bla bla, as every modern ssd, even consumer ones, does have this. It just ensures that data already written to NAND doe not become corrupted whilst power failure, so just a feature of the firmware.
Both your Intel and your Micron does have this, but Micron makes it a marketing thing.

see for instance here:
Protecting Data Integrity During Power Loss Events

 

[amalurk]

Not sure but it sounds like marketing speak where you make up a similar name to a real feature to pretend you have something similar when you actually don't have the real feature.

 

[RolloZ170]

there is a mn. hold time required in ATX power supplies, after AC loss power good signal is negotiated and gives time to flush the SSD cache, but i don't know if this is implemented in consumer motherboards. better let the SSD do the job itself thought.

 

[CyklonDX]

On desktops, it depends on the system load, and PSU wattage rating.
The really expensive 1kW titanium seasonic can hold for around 2sec without power when it was under ~200W load.
(but only due to capacitor load, discharge ratio.) All to do with caps capacity on psu. In terms of servers, its mainly not for flush of caches; but time for gate switch to react for one of psu going down, or power going off on one of them.

 

[unwind-protect]

How would the SSD know that the PSU lost power and go from regular operations to flushing buffers?

 

[RolloZ170]

How would the SSD know that the PSU lost power and go from regular operations to flushing buffers?

PowerGood from PSU, this will result in reset going low.

 

[CyklonDX]

How would the SSD know that the PSU lost power and go from regular operations to flushing buffers?

The controller is always trying to flush its internal caches to permanent storage. Once system looses power, and no new write signals are recived its able to flush its buffers very quickly. In short the disk doesn't know it lost power (it does - by sending flush command to its cache), and just operates until power is depleted.

 

[Stephan]

Those yellow caps are imho ceramic. Only good to prevent the worst of outcomes like half written pages imho. What you want is the black caps of type electrolyte. See Micron 5100 MAX & ECO Enterprise SSD Review (960GB/1920GB) - The SSD Review (scroll a little) These hold enough juice to transfer in-flight data to a microcontroller for persistence which has enough FERAM or similar and works with lower voltage than main flash controller like 1.8V so can hold out longer. When the SSD is receiving good power again, the data is recovered and written to NAND.

 

[111alan]

The capacitors are used to offer PLP for „data at flight“, implemented for most enterprise / datacenter ssd‘s. This implementation makes sure that data sitting in the (RAM)Cache becomes flushed to the NAND in case of power failure.
This is often referred as „end to end“ PLP.
This is implemented in your Intel ssd, thus you can see the capacitors value within SMART values. The Micron 1100 does NOT have this kind of PLP implemented, thus no data for capacitors in SMART.

For „data at rest“ PLP is more marketing bla bla, as every modern ssd, even consumer ones, does have this. It just ensures that data already written to NAND doe not become corrupted whilst power failure, so just a feature of the firmware.
Both your Intel and your Micron does have this, but Micron makes it a marketing thing.

see for instance here:
Protecting Data Integrity During Power Loss Events

"Data at rest" should refer to "lower/middle page". The corruption happens like this:

For example, like, when we're talking about a TLC NAND flash, for each cell it can hold three bits. Then a group of these cells can form three "pages", where each page was made from of one of three bits of data from each cell. Hence "lower" "middle" and "upper" page(or "lower""upper"and"extra" depending on manufacturer's definition). When there's already data in the "lower" page, if the SSD want to write the"middle" page, it has to tamper with the same cells which stores the "lower" page data.

If the power loss happens right in the time period when the NAND is writing the middle or upper page, the operation will be incomplete, leaving the cells at the wrong voltage which are most likely won't be representing any correct data. Then the data will be corrupted no matter what actions you take afterwards.

For enterprise drives, this is not a problem, since all write operations will be completed during SPOR, there will be enough power stored in the capacitors to do that(unless there's some sever design flaws). For consumer drives, what vendors were saying about the protection is, the drives would most likely not to kill itself during power loss, and when power was resumed, the FTL can be recovered from a backup, so that the drive can still live. Data is most likely corrupted in this case whatever. If you happend to wrongly unplug the system drive or an SSD in a USB enclosure, you may know what I'm talking about,

Just saying, all the protections in consumer drives are aiming for protecting the drive from dying outright, while only the enterprise drives have any real concerns for data integrity.

 

[unwind-protect]

Just saying, all the protections in consumer drives are aiming for protecting the drive from dying outright, while only the enterprise drives have any real concerns for data integrity.

Thank you for that description. Could you shed some more light on the difference between "full" PLP and PLP-at-rest (if any difference)?

The wording makes it sound like there is less capacitor capacity and it would only preserve the layers that previously existed, but not flush any new data.

 

[111alan]

Thank you for that description. Could you shed some more light on the difference between "full" PLP and PLP-at-rest (if any difference)?

The wording makes it sound like there is less capacitor capacity and it would only preserve the layers that previously existed, but not flush any new data.

Just checked some documents. Seems that this is one of the rare exceptions that a consumer SSD is actually trying to protect lower page faults. The capacitors can hold power for 1ms, during which the cells being written can be reverted back to the state before the write operation.

This method will still discard all the on-flight data as well as changes in FTL which weren't backed up to the NAND.

In the case of “full” PLP, nothing will be lost. Write operations will be allowed to complete, FTL and data buffer inside the DRAM or SRAM cache will be flushed to the NAND as well.

https://www.micron.com/-/media/client/global/documents/products/white-paper/ssd_power_loss_protection_white_paper_lo.pdf

 

[Whaaat]

micron uses ceramic capacitors on their disks.

Those capacitors typically provide from 1ms to 50ms of uptime designed to last enough time to flush the data from on-disk cache/ram to nand modules.

SSD (10-50ms)
View attachment 28679

The pack of plp capacitors inside Samsung's SM1625 does look like the drive is not going to cease its operation during the entire blackout

 

[CyklonDX]

The pack of plp capacitors inside Samsung's SM1625 does look like the drive is not going to cease its operation during the entire blackout

View attachment 28774

are those gum packs? I mean you don't get that much even on raid controllers. This looks like it'll be able to keep caches an hour or so.

 

[unwind-protect]

are those gum packs? I mean you don't get that much even on raid controllers. This looks like it'll be able to keep caches an hour or so.

I speculate that the SSD they are sitting on might also be a power hog.

 

[Whaaat]

I speculate that the SSD they are sitting on might also be a power hog.

Most of SAS ssd's are capped at about 10W while writing

are those gum packs?

at first I thought these are the packs of c4 explosive planted there for the sake of 'instant secure erase' once and for all

 

[CyklonDX]

at first I thought these are the packs of c4 explosive planted there for the sake of 'instant secure erase' once and for all