MX100 power loss protection is "fake" ?

[Elixer]

http://www.anandtech.com/show/8528/micron-m600-128gb-256gb-1tb-ssd-review-nda-placeholder

In the MX100 review, I was still under the impression that there was full power-loss protection in the drive, but my impression was wrong. The client-level implementation only guarantees that data-at-rest is protected, meaning that any in-flight data will be lost, including the user data in the DRAM buffer. In other words the M500, M550 and MX100 do not have power-loss protection -- what they have is circuitry that protects against corruption of existing data in the case of a power-loss.

Hmm, that makes me think twice about the MX100. Saying it has one thing, when it doesn't really have power loss protection paints a black eye on Crucial's specs of the MX100.

 

[KingFatty]

Is there a way to configure windows to address this? Maybe change the setting about how data caching is write ahead or whatever?

 

[BonzaiDuck]

http://www.anandtech.com/show/8528/micron-m600-128gb-256gb-1tb-ssd-review-nda-placeholder

Hmm, that makes me think twice about the MX100. Saying it has one thing, when it doesn't really have power loss protection paints a black eye on Crucial's specs of the MX100.

Step into my Time Machine for a trip back to 2007.

Crucial Ballistix and Tracer memory modules were widely touted and popular. They were deemed "very overclockable," or capable of running at tighter latencies. I was one of those folks who volted the modules higher but STILL WITHIN THE WARRANTY-SPEC.

I don't know if there was a discussion about it, but I went through one kit of Ballistix and two kits of Tracers. (These latter were rebadged as "LanFest" as part of a Newegg promotion -- but they were Tracers plain and simple -- with timings and voltage identical and the same LED bling.)

I don't think I even bothered to RMA the Tracers. I DID RMA the Ballistix kit, and the replacement did not run at the full spec. I concluded that Crucial had been too ambitious when they spec'd these kits. There were indications that Crucial's RMA division was undergoing a crisis at that time.

On this particular matter of SSDs and "power loss protection," it isn't exactly a feature being touted by Crucial's competitors. I don't think their SSD warranty is much different than any other. I've been satisfied with my two MX100's for a few months now. Only time will tell . . .

And just another point. If you're worried about power-loss leading to data-loss, you should be running your systems off UPS backups. If your system is unstable for some reason, that's something you should've tested when you first built it, and the instability will cause crashes (with possible data loss) with or without battery backup.

Eventually, there is a chance that some part will fail before you're ready to retire a PC (or laptop . . . or tablet . . . these SSDs have many uses). There is always some chance of "data-loss." I say -- back up the data, and back up the power . . .

 

[Hellhammer]

Is there a way to configure windows to address this? Maybe change the setting about how data caching is write ahead or whatever?

Yes. You can disable the use of DRAM buffering in Device Manager (right-click the drive -> Properties -> Policies -> Uncheck the box). However, there may be a loss in write performance and even that doesn't ensure data protection since any ongoing NAND writes will still be lost.

 

[Elixer]

And just another point. If you're worried about power-loss leading to data-loss, you should be running your systems off UPS backups. If your system is unstable for some reason, that's something you should've tested when you first built it, and the instability will cause crashes (with possible data loss) with or without battery backup.

Eventually, there is a chance that some part will fail before you're ready to retire a PC (or laptop . . . or tablet . . . these SSDs have many uses). There is always some chance of "data-loss." I say -- back up the data, and back up the power . . .

Yeah, I know that, and, in fact I do have a UPS and do backups.

I was just stating that it is a "dirty trick" saying your product does one thing, when it doesn't do that.
Then NOT correcting reviewers when they wrongly assume it has full power loss protection.
You know Crucial would be all over the reviewer if they got something else wrong, but, in this case, they chose to look the other way, when they knew what was stated was false.

 

[Mark R]

Is there a way to configure windows to address this? Maybe change the setting about how data caching is write ahead or whatever?

Windows file systems (including FAT) will protect the file-system metadata in the event of a power-failure during writing. In NTFS, this is done with a journal. In FAT, this is done by using a specific sequence of writes, allowing an incomplete write to be detected at bootup, and then having scandisk rollback the incomplete metadata write.

Windows doesn't attempt to protect user data in this way, in any file system (I don't think even ReFS attempts this).

As a result of this, it is recognised that power loss while saving a file can result in a corrupted file - many apps, notably MS office, therefore save a new file with a temp file name, then once that is saved, they delete the old file, and rename the temp file.

Windows, in general, always keeps a write back buffer for non-removable drives, and I don't think this can be disabled. There is an option in device manager to disable a drive's internal write cache (usually reserved for drives with buggy caches which might corrupt metadata, or for removable drives).

In general, not having power loss protection isn't really that big a deal. Provided that the drive's cache works correctly, there shouldn't be any major problems. The risk is that the cache doesn't correctly honor write barriers (a signal from the OS which tells the drive, all pending write before the barrier must be safe, before any pending writes after the barrier are allowed to be safe). This is a huge problem for most OSs if write barriers are ignored.

One study of SSD reliability showed that almost all SSDs would ignore write barriers in a power-loss situation, even those with power-loss protection. Correct cache behaviour under power failure conditions is difficult to test, but this is the key feature to look for in enterprise level drives.

 

[BonzaiDuck]

I guess the nagging question goes something like this. If Crucial "overly hyped" the power-loss correction feature, how can you trust anything else about the product? Well -- the benchmark programs on the MX100 seem to verify that it runs to spec.

 

[Racan]

The Intel 730 is the only client SSD with true power loss protection.

 

[meloz]

http://www.anandtech.com/show/8528/micron-m600-128gb-256gb-1tb-ssd-review-nda-placeholder

In the MX100 review, I was still under the impression that there was full power-loss protection in the drive, but my impression was wrong. The client-level implementation only guarantees that data-at-rest is protected, meaning that any in-flight data will be lost, including the user data in the DRAM buffer. In other words the M500, M550 and MX100 do not have power-loss protection -- what they have is circuitry that protects against corruption of existing data in the case of a power-loss.

Sorry, but how is this any different from every other SSD in the market, particularly the popular Samsung ones? Do Samsung 840, 850 et al corrupt stored data on disk in the event of a power loss? In that case Samsung should get into UPS business, give consumers bundle deals with their SSDs.

And why does no one test SSDs for this? These SSD reviews have all become such shallow marketing exercises. It would be easy to test a SSD's resilience to power loss, pull the plug on a PC while running a game or playing a movie. Do it ten times or so and see what happens.

 

[Elixer]

Sorry, but how is this any different from every other SSD in the market, particularly the popular Samsung ones? Do Samsung 840, 850 et al corrupt stored data on disk in the event of a power loss? In that case Samsung should get into UPS business, give consumers bundle deals with their SSDs.

And why does no one test SSDs for this? These SSD reviews have all become such shallow marketing exercises. It would be easy to test a SSD's resilience to power loss, pull the plug on a PC while running a game or playing a movie. Do it ten times or so and see what happens.

Read http://forums.anandtech.com/showpost.php?p=36759104&postcount=5

They never should have stated it the way they stated it, since it wasn't the truth.

 

[Hellhammer]

And why does no one test SSDs for this? These SSD reviews have all become such shallow marketing exercises. It would be easy to test a SSD's resilience to power loss, pull the plug on a PC while running a game or playing a movie. Do it ten times or so and see what happens.

I can tell you what happens: nothing. If you're gaming or playing a movie, there is no write activity going on, so there is no ongoing writes to be lost.

The thing is that client SSDs do not have power-loss protection (Micron is the only one that claimed to have it but that wasn't exactly correct). I don't see why testing something that isn't there in the first place is something we should do.

Besides, client-grade hard drives never had power-loss protection either and inherently suffered from the same issue. That is why all moderns OSs are built to sustain minor data losses prior to the sudden shut down

 

[Mark R]

Besides, client-grade hard drives never had power-loss protection either and inherently suffered from the same issue. That is why all moderns OSs are built to sustain minor data losses prior to the sudden shut down

It's worse than that though. Many client-grade SSDs are liable to data corruption, either through corruption of internal metadata, or through failure to honor write barriers under power-loss conditions (therefore defeating the data-loss prevention systems in the OS).

It appears that the MX100 "power-loss protection" system merely means that the drive shows the "expected" power-loss behviour.

 

[Hellhammer]

It's worse than that though. Many client-grade SSDs are liable to data corruption, either through corruption of internal metadata, or through failure to honor write barriers under power-loss conditions (therefore defeating the data-loss prevention systems in the OS).

It appears that the MX100 "power-loss protection" system merely means that the drive shows the "expected" power-loss behviour.

The corruption of the flash translation used to be an issue but now it seems that most players have it figured out by using journaling and/or issuing more flush commands to flush the FTL from DRAM to NAND. At least I rarely see any failures that appear to be caused by FTL corruption anymore -- there used to be a lot more a year or two back.

 

[Cerb]

When was it stated that the M500 and newer's power loss protection was the same as enterprise drives? I was under the impression from the start that it was added protection against bricking, or non-bricking metadata corruption, since dropping voltages have a nasty habit of scrambling data before devices actually shut down. Micron's own docs have made it pretty clear that it's not good enough for committing all pending writes.

While a decent drive for its day, the M4 could suffer just that sort of corruption.

The image in question is marketing. It's stating the truth, but in a way that lends itself to misinterpretation. WD has been doing the same for a long time. Don't trust any marketing blurbs to be plainly explaining anything.

 

[Hellhammer]

Micron's own docs have made it pretty clear that it's not good enough for committing all pending writes.

Take a look at the snippet from MX100's product page that was posted earlier in this thread. It, if something, is giving the picture that pending writes would be completed in the case of a power-loss.

 

[Cerb]

Take a look at the snippet from MX100's product page that was posted earlier in this thread. It, if something, is giving the picture that pending writes would be completed in the case of a power-loss.

Just like Intellipower's marketing-speak gave the impression of variable RPMs on WD Greens. Reading the product page for the Purple, recently, I found my eyes could only roll so far...

IMO, almost every computer technology company's marketing deserves to be called out, or ignored completely, because the handful that aren't so severely skewing the truth are those that just can't, and those that skip consumers as direct buyers (Intel confuses more than compels with their marketing, and Toshiba doesn't even try, FI).

In the M500's public data sheet's cover page:
"Power loss protection for data-at-rest"

The M550 and M600 have the same thing.

But, :hmm:, I cannot find such a data sheet for the MX100. It's all purely marketing materials.

 

[Hellhammer]

But, :hmm:, I cannot find such a data sheet for the MX100. It's all purely marketing materials.

And that's part of the problem. Crucial is the retail side of the business, so we and the consumers deal with that side. I wouldn't have a problem if Micron's and Crucial's materials were consistent with the way they refer to PLP, but they aren't.

The above is taken from the data sheet Crucial sent me for the MX100 review. It's just PLP there with zero mention of it being limited to data-at-rest. I guess the Micron side is better since they deal with OEMs, but Crucial's materials are still misleading since generally PLP refers to its full implementation. There should be a clear distinction between client and enterprise-level implementations -- both can't be just called power-loss protection.

I do see your point about marketing and as I said in the review, marketing will and never should be fully honest and thorough. However, I had to call them out because they got me with their marketing, so I had been promoting a feature that was never there.

 

[Topweasel]

This is from Crucials MX100 main page.

http://www.crucial.com/usa/en/storage-ssd-mx100

Protect all of your data - even if the power goes out.

With Power Loss Protection, if you happen to be saving a file and you suddenly lose access to power, your data will still be preserved.**

At the vary bottom of the Crucial Mx100 main page.

Note: All trademarks are property of their respective owners. 1 GB = 1 billion bytes. Actual usable capacity may vary.
* Hard drive active average power use based on published specs of 1TB Western Digital® Blue WD10JPVX internal hard drive. According to NPD data published February 2014, this was one of the most popular hard drives available and an accurate reflection of a common internal hard drive.
**Refers to data at rest only
***Performance level based on comparative AS SSD benchmark ratings of 512GB Crucial MX100 SSD and 1TB Western Digital® Blue 7200 RPM hard drive (WD10EZEX-00U). Actual performance levels may vary based on benchmark used and individual system configuration. Test setup: Intel® DZ87KLT-75K motherboard, Intel i7-4770K 3.50GHz processor, BIOS Rev. 0446, and Windows® 7 Ultimate 64-bit operating system using AS SSD test suite.

Sure they use stupid footnotes to clarify their marketing speak but its the same thing everyone does.

 

[Coup27]

You should not be allowed to say "Protect all of your data" but then a footnote saying "Not all of your data is protected".

Crucial / Micron have definitely pulled a fast one with this. Nobody from them felt the need to jump in and correct any tech articles when they were promoting this feature originally.

 

[Elixer]

This is from Crucials MX100 main page.

http://www.crucial.com/usa/en/storage-ssd-mx100

At the vary bottom of the Crucial Mx100 main page.

Sure they use stupid footnotes to clarify their marketing speak but its the same thing everyone does.

Well, they finally fixed it then.

Notice the image that I had posted, and now, they added the ** footnote.


As you can clearly see here: http://web.archive.org/web/20140902084159/http://www.crucial.com/usa/en/storage-ssd-mx100

Not one mention of "data at rest only".
That means they where indeed trying to pull a fast one, and hope that nobody would notice, and that worked for the most part, until recently.

 

[C101]

In my industry, power loss protection means the drive should be able to withstand everything up to and including a crowbar of the input power lines without any data loss.

I have yet to see a consumer SSD that is capable of this, though some look like they have the bulk capacitance on-board to support a hot unplug of the drive. The Intel SSD mentioned earlier is a good example of this, with the two large electrolytic capacitors on-board.

As an aside, I'd like to see pictures of both sides of the SSD PCBA in reviews. Too many only show one side, and that can sometimes hide

 

[ashetos]

Setting the misleading marketing aspect aside, data-at-rest protection is good enough since it does not break file-system semantics and lets the journals do their thing.

But I still wonder what they mean with the term data-at-rest protection. It should mean that every write command that has sent a completion to the O/S driver is considered data at rest.

But that means they also have to protect data that move from the SLC cache to the MLC storage or other fancy stuff like that. I hope this is what data-at-rest protection means.

 

[Cerb]

But I still wonder what they mean with the term data-at-rest protection. It should mean that every write command that has sent a completion to the O/S driver is considered data at rest.

Data not being written, physically or electrically nearby on the chip, could get corrupted during power loss, with writing going on during power loss. Developments in the NAND itself have been ongoing to deal with this problem, as well as work in the SSD controllers (IMFT have been vocal about such work, but surely everyone else is doing the same).

HDDs are not immune to similar (but not identical) issues, with unrelated data getting scrambled. But, HDDs don't risk bricking or total drive corruption from such an event (though hosing a partition can happen, however rarely).

When power is lost, voltages don't fall at the same rate across every wire. Operations currently going on can continue, but wrongly, if one or more high voltages turns to low during that window. So, you might want either a way to tag a write going on as bad, or a way to keep whatever it's doing that might go wrong isolated. If there's not enough power for the complete write, as long as it can reliably be rolled back to the state before that write began, then it's up to the file system to handle the rest.

So, yes, it should mean that if the SATA host controller got a success status back, then it aught to be protected. Generically, power loss protection means that if the drive acknowledged receiving the write command, then pulling power right then should still result in a successful write. So, their M-series drives specifying that it is not for all in-flight data is an important piece of information.

In higher-end servers, it has not been uncommon for RAID cards, and sometimes high-end HDDs themselves, to have a battery backup on them, so that any writes sent have time to complete. That way, the drives are all in an idle state during power off, even if the power off was due to the machine itself crashing, rather than a power outage. It's also done that way because many server systems handle remote information, and those remote nodes need to be assured that any sent data they got a successful ack for was actually committed (and thus why a UPS for the whole computer is not sufficient). Consumer systems, and most SMB systems, have none of this, and tend not to need it, either.

But that means they also have to protect data that move from the SLC cache to the MLC storage or other fancy stuff like that. I hope this is what data-at-rest protection means.

Making everything transactional would handle that fine, so long as there's no risk of screwing up a whole block, unknowingly leaving a half-written page, screwing up another page on another block, etc. etc.. IE, the data isn't moved, but copied, verified, then the pSLC portion marked as re-usable. Not any different than moving around plain MLC, which the drives all have to do regularly.

I have yet to see a consumer SSD that is capable of this, though some look like they have the bulk capacitance on-board to support a hot unplug of the drive.

A small handful of Intels are pretty much it, just because they share hardware w/ their enterprise market brethren. FI, the 730 from Intel is technically a consumer drive.