Questions about Power Loss to SSD's

[aman74]

I'm sure this is old news around here, but I'm doing a new build, possibly Hackintosh and haven't done one in over 10 years so am researching and updating my knowledge. I hadn't heard about this one before, so maybe it's not a big deal, but right now I'm concerned.

Came across this article: http://www.extremetech.com/computin...ing-drive-are-power-outages-killing-your-ssds

-What is PLC and MCL?

-I see some drives have "power loss protection" listed as a feature, but have heard reference to it not doing anything? Any drives actually protect against this?

In the comments section to that article someone asks this:

"Does this only happen when the drive loses power during a data transfer? For instance, if my system locks up and I have to do a hard restart, would disconnecting the data cable (no data cable, no read/writes?) and then doing the hard restart, save the drive?"

The author replies:

"No, it wouldn't. But a hard fault may be treated differently than a lockup. That's one of those ugly questions I don't have an answer to."

Then someone quotes the research that the article references, which seems to suggest that the issue would be prevented if you removed the data connection? Suggest the authors reply was incorrect? However, I may be misunderstanding.

Here's the research quote:

"Just power cycling the drives while no read/write operations were occurring was no problem, but power cycling them during the read-synchronize-write cycle was incredibly problematic. After 1600 power cycles, the M4 was recording up to 40,000 CRC errors. The Toshiba THNSNH060GCS upgrade kit was able to maintain file integrity if file writes were handled at less than 20MB/sec total, even when writing 64 threads of data. Exceed that rate, however, and the Toshiba drive starts losing data quickly."

Unfortunately the author never replied to that.

Also, what is the difference between a "hard fault" and a "lockup" that the author references in his reply?

Thanks much!

 

[CiPHER]

• All SSDs need some kind of protection against corruption of the mapping tables; or you will have massive failures like in the early days when SSDs were without protection and would corrupt themselves easily, bricking the drive in the process.
• Power-safe capacitors is one way to protect against inconsistent mapping tables; it is the best way though. But Samsung doesn't use any hardware protection but instead uses a software protection: journaling for the mapping tables. Your SSD will revert itself to an earlier state which Samsung calls POR (Power-On Recovery). This is okay for desktop usage but not okay for anything else like when using the SSD in RAID arrays.
• Any power interruption without first sending STANDBY IMMEDIATE ata-command will count as unexpected power-loss for the SSD. This includes software crashes, power-cycles outside of the operating system and other failures other than traditional power-failure. Using an UPS does not negate the need for protection against unexpected power-loss.
• Unprotected power-loss may damage the NAND, spoil the writes still in buffercache, but worse: it will make the mapping tables inconsistent with the stored data.
• What the author says about CRC makes no sense to me; CRC is only used for detecting corruption related to the SATA interface (the cable). Not CRC, but ECC is used to protect the NAND against bitrot. ECC means error correction, while CRC can only detect errors, and not correct them.

So the short story is that you need some kind of mechanism to cope with the inherent flaws of NAND, and that some manufacturers choose a different path. Like Samsung which saves some money and uses a software implementation, while Micron (Crucial) uses a hardware protection which is generally superior to the software implementation. For desktop users, both are equal i'd say.

 

[aman74]

Thanks for that...quite a bit over my head, but I'll try to take it in.

You mention RAID...so if I ever wanted to RAID two SSD's I need to get a certain kind of drive?

You mention BIT Rot. I just came across something about that the other day. I assume all SSD's are ECC? Is ECC system RAM somehow more important for SSD's?

I read something about new file systems that help with bit rot. Are these in use yet? What file system would I use for a hackintosh?

I've read there were issues with the MX100, but haven't looked up yet what they are? The other two drives that seem to get the most mentions are the Sandisk and the Samsungs, but the 840 series had issues. Are those corrected in the new 840's?

This seems like a pretty serious issue and would like to mitigate it as much as possible.

 

[CiPHER]

You mention RAID...so if I ever wanted to RAID two SSD's I need to get a certain kind of drive?

Generally all SSDs will do, but some SSDs like the Samsung use some special techniques that could under some circumstances behave erratic in a RAID array or other advanced storage.

Just imagine having four SSDs in RAID0 and two of them revert to an older state thanks to the 'Power-on Recovery'. Since one filesystem is spanned across all four SSDs, now parts of the filesystem are in a much older state than the other parts. If all SSDs would revert into the past this would not be such of a problem, but now it will mean the filesystem will check itself and correct 'damage' that actually is just inconsistency.

So if you want to RAID them, i would not recommend any Samsung SSD.

You mention BIT Rot. I just came across something about that the other day. I assume all SSD's are ECC? Is ECC system RAM somehow more important for SSD's?

All harddrives use ECC, all SSDs use ECC. Without ECC they would be like alzheimer and only remember a part of their data. ECC is mandatory for storage devices, but not for your system RAM. There is no technical reason not to provide ECC-protection to your RAM as well, but the manufacturers do not want this and instead generate additional profit from keeping ECC a feature enterprise users have to pay the big bucks for.

To prevent bitrot, modern SSDs use three protections:

• ECC prevents unreadable NAND pages ('sectors') comparable to a bad sector on a harddrive.
• Internal RAID5 parity prevents bitrot even when the ECC fails and a page becomes bad. Yes, SSDs actually use RAID3 or RAID5 internally. More primitive SSDs use RAID0 (interleaving) without parity. This is performed internally by the SSD controller; not by the host. If you put two SSDs in RAID0, each with 16-way interleaving, then you could say you have a 32-way interleaved storage device. Otherwise put: 32 NAND dies in RAID0. The commercial name for this is 'RAIN' (Micron) or 'RAISE' (Sandforce) whose name obviously was inspired by RAID.
• The power-safe capacitors or journaling make sure the mapping tables (the 'index' of what is stored and where) match the stored data. Harddrives and simple flash devices lack mapping tables and thus would not need to protect them. Without mapping tables (prior to Intel X25-M era) the SSDs were very slow in random writes, slower than harddrives. With the mapping tables, they are up to 100 times faster than harddrives. So it makes a big difference, but adds complexity that can cause failure. Today this complexity has been managed pretty well. But it should be said that all major SSD brands had serious dataloss bugs over the years and the cause for this is the complexity of modern (fast) SSDs.

I read something about new file systems that help with bit rot. Are these in use yet? What file system would I use for a hackintosh?

Sorry but Apple, like Microsoft, is very much behind in terms of storage. Apple OSX uses HFS filesystem and it generally at the same level of NTFS: good for the year 2000 but by todays standards outdated and obsolete.

It'll work just fine though, and you don't really have a choice. ZFS was once in OSX (version 9 of ZFS i believe) but it has been removed after Oracle acquired Sun, probably from the fear of reprisals from Oracle. Which is not that strange as it's a very evil company.

I've read there were issues with the MX100, but haven't looked up yet what they are?

I'm not aware of any issues with the MX100. It's generally the best consumer SSD one can buy. The MX200 will be a bit better though, as it offers dynamic write technology. It allows you to use the SSD as an SLC SSD if you use up to 50% storage, and it also accelerates writes by writing it to SLC instead of MLC.

This feature is basically an improved version of the dynamic write technology in the Samsung 840/850 EVO series. But unlike Samsung, Crucial is shipping a 'proper' MLC drive while Samsung gives you TLC. And the Samsung is actually more expensive, from a technical standpoint it should be much cheaper than the Crucial. But it's actually the other way around.

The other two drives that seem to get the most mentions are the Sandisk and the Samsungs, but the 840 series had issues. Are those corrected in the new 840's?

The 840 EVO has performance drops, and they are not fully fixed. The 850 EVO is unaffected as far as i know. But these issues are rather minor; performance differences between SSD are measurable but hardly noticeable. In fact, it is the 20MB/s random 4K read performance that makes SSD feel so snappy and fast compared to a harddrive, not their high 500MB/s sequential speeds. Those are only 2 tot 3 times as fast as a harddrive, while an SSD is 100 to 1000 times faster in terms of random I/O performance.

 

[Cerb]

-What is PLC and MCL?

???

-I see some drives have "power loss protection" listed as a feature, but have heard reference to it not doing anything? Any drives actually protect against this?

When power is lost, what stored energy there is, as seen as voltage, drains faster along some paths than others. It's not nice and even. So, if a write is going on, it may go try to write to another location, or write corrupted data, while the voltage is dropping, but hasn't reached a level to shut it off, yet, because the same voltage that is on those power lines is also used for determining what's a 0 or 1. So, some methods need to be used to cope with that. At the least, the SSD needs to be able to either complete any current writes, or interrupt them, safely. Then, it needs to be able to detect that voltage is dropping quickly from the PSU, so that it can get that, and any other work it needs to do, done, before local power already stored on the device is lost.

Intel/Micron and Sandisk/Toshiba have touted enhancing the flash chips themselves to be resistant to such issues, generation over generation (details, sadly, are competitive advantages, AKA secrets). Due to that, the M500 and M550 have capacitors that give them a mere single microsecond, and that is sufficient; even though actually programming flash might need over a millisecond (I haven't looked at the M600's docs regarding that, and the consumer drives lack good public documentation).

Also, what is the difference between a "hard fault" and a "lockup" that the author references in his reply?

Think power outage v. BSOD.

You mention BIT Rot. I just came across something about that the other day. I assume all SSD's are ECC? Is ECC system RAM somehow more important for SSD's?

It's important to everything, but price and market segmentation are too important to get it in our desktop DRAM. Basically everything but PC and phone/tablet/console DRAM gets plenty of ECC. As small as bits are in DVDs, BDs, HDD platters, and on chips, you can assume that most of the time, the data was written with errors from the outset. The effective (literal for HDDs) SnR has gotten very low, and it's easier to cram more bits in the space to use for data correction than to improve the signal quality, or slow down reading and writing enough to be able to perfectly read or write with so much 'noise'.

It's nothing new or unusual. The audio CD has ECC, though not to the level of any data storage medium (some loss was considered acceptable), and that's from way back in the mid/late 70s.

 

[aman74]

Sorry but Apple, like Microsoft, is very much behind in terms of storage. Apple OSX uses HFS filesystem and it generally at the same level of NTFS: good for the year 2000 but by todays standards outdated and obsolete.

It'll work just fine though, and you don't really have a choice. ZFS was once in OSX (version 9 of ZFS i believe) but it has been removed after Oracle acquired Sun, probably from the fear of reprisals from Oracle. Which is not that strange as it's a very evil company.

I'm not aware of any issues with the MX100. It's generally the best consumer SSD one can buy. The MX200 will be a bit better though, as it offers dynamic write technology. It allows you to use the SSD as an SLC SSD if you use up to 50% storage, and it also accelerates writes by writing it to SLC instead of MLC.

This feature is basically an improved version of the dynamic write technology in the Samsung 840/850 EVO series. But unlike Samsung, Crucial is shipping a 'proper' MLC drive while Samsung gives you TLC. And the Samsung is actually more expensive, from a technical standpoint it should be much cheaper than the Crucial. But it's actually the other way around.

Thanks again...much of it over my head, but it helps a lot and I'm sure will help others as well as they see the thread and even find it some time later in searches...

I saw reference to problems with the MX100, but the post didn't give details. I'll have to see what google reveals, etc... I have heard it's highly regarded, but the Samsung seems more popular. I'm curious about their RAPID technology as well. I know they all feel snappy, but when you're using one for a scratch disk and/or running out of RAM in photoshop, I'd think as close to RAM speed as you can get it would benefit. I started to research the M.2 and PCIe drives and man what a rabbit whole that is as far as figuring out what path the MB uses beyond the physical connector. Same as it ever was in computers and evolving standards.

I saw mention so people using NTFS with Apples...didn't know it was possible, but now I have to look into that as well.

The article I saw talking about bit rot and newer or maybe just alternative file systems was probably talking about enterprise use or something and I didn't realize it was something I couldn't use.

???

When power is lost, what stored energy there is, as seen as voltage, drains faster along some paths than others. It's not nice and even. So, if a write is going on, it may go try to write to another location, or write corrupted data, while the voltage is dropping, but hasn't reached a level to shut it off, yet, because the same voltage that is on those power lines is also used for determining what's a 0 or 1. So, some methods need to be used to cope with that. At the least, the SSD needs to be able to either complete any current writes, or interrupt them, safely. Then, it needs to be able to detect that voltage is dropping quickly from the PSU, so that it can get that, and any other work it needs to do, done, before local power already stored on the device is lost.

Intel/Micron and Sandisk/Toshiba have touted enhancing the flash chips themselves to be resistant to such issues, generation over generation (details, sadly, are competitive advantages, AKA secrets). Due to that, the M500 and M550 have capacitors that give them a mere single microsecond, and that is sufficient; even though actually programming flash might need over a millisecond (I haven't looked at the M600's docs regarding that, and the consumer drives lack good public documentation).

Think power outage v. BSOD.

It's important to everything, but price and market segmentation are too important to get it in our desktop DRAM. Basically everything but PC and phone/tablet/console DRAM gets plenty of ECC. As small as bits are in DVDs, BDs, HDD platters, and on chips, you can assume that most of the time, the data was written with errors from the outset. The effective (literal for HDDs) SnR has gotten very low, and it's easier to cram more bits in the space to use for data correction than to improve the signal quality, or slow down reading and writing enough to be able to perfectly read or write with so much 'noise'.

It's nothing new or unusual. The audio CD has ECC, though not to the level of any data storage medium (some loss was considered acceptable), and that's from way back in the mid/late 70s.

???....not sure what the question is? They are mentioned in the article I linked to.

Is are the Crucial drives basically the new Intel drives then? I remember in the early days of SSD the Intel drives were considered the safest. I haven't seen them marketed lately, so wondering if it's now just the Crucial brand that is licensing and marketing the technology? Crucial is Micron and is not just a rebrander like OCZ or Corsair correct? They make their own RAM and SSD's?

Thanks for all the info as well. Much over my head as well, but I'm learning.

 

[aman74]

Also wanted to ask again...do the drives that say they feature power loss protection actually have any advantage?

I found an update to that article I linked:

http://www.extremetech.com/computin...might-be-the-only-reliable-drive-manufacturer

So this brings me back to my biggest question from my OP.

I gave the quotes from the original comments section about the drive failing during I/O operations. The author had replied to someone's question about unplugging the data cable if you have a lockup and then forcing the hard boot. He seemed to say that wouldn't help. The quote from the research article someone posted as a follow-up seems to directly contradict this and the author never came back to reply.

Now this new updated article seems to clear this up somewhat or at least to my understanding. It says the issue isn't from sudden power loss, but sudden power loss during read/writes.

Am I correct?

Honestly, I very much appreciate all the technical info, but either I was answered about this and it's just over my head or my original question got a bit lost there somewhere.

I'm just trying to find out the best practical ways to go about mitigating this and if any drives would be better at this then others.

It does appear that a UPS is a good idea and also that if you did have a lockup that was unable to be resolved that disconnecting the data cable before a hard boot would eliminate the problem?

I found a deal on a Samsung and am thinking of pulling the trigger, but wanted to find out of the drive matters much as far as this issue is concerned first. The newer article indicated Intel was the only one that did well in the test, but that was only two older and smaller drives, so who knows if things have changed.

 

[CiPHER]

Also wanted to ask again...do the drives that say they feature power loss protection actually have any advantage?

I thought i had explained that in detail?

Power-loss protection is crucial; if the SSD doesn't have it, it's either a low-grade SSD or it needs another way to protect its sensitive mapping tables, like Samsung does with FTL journaling.

Now this new updated article seems to clear this up somewhat or at least to my understanding. It says the issue isn't from sudden power loss, but sudden power loss during read/writes.

The SSD can be busy doing background I/O so it is possible the SSD is active even though you are not using it at all.

Honestly, I very much appreciate all the technical info, but either I was answered about this and it's just over my head or my original question got a bit lost there somewhere.

Then let's make it simple:

No protection at all: garbage SSD (bad)
sofware protection: Samsung (mediocre)
hardware protection: Crucial (best)

It does appear that a UPS is a good idea

Always a 'good idea' but not that relevant to SSD power-loss protection as you need that anyway, even if you have an UPS.

 

[aman74]

I thought i had explained that in detail?

Power-loss protection is crucial; if the SSD doesn't have it, it's either a low-grade SSD or it needs another way to protect its sensitive mapping tables, like Samsung does with FTL journaling.

No, not really, at least in the context of the way I asked it. Very few are advertising some sort of power protection and I believe the Crucial M500 was the first to advertise it and people were claiming it wasn't "real"...that part I'll have to look into.

The SSD can be busy doing background I/O so it is possible the SSD is active even though you are not using it at all.

Yes, of course, but that's not the point nor the question. Let's make this simple. I didn't ask whether it mattered if I was using it at all or thought it wasn't doing a read or write I asked if the major issues are when the drive is active. So after all this I still get no answer.

If the computer locks up and hasn't lost power and there's no harm to an SSD from a hard boot if it's not active, then pulling the cable would be of great benefit so you could the hard boot at that point.

Then let's make it simple:

No protection at all: garbage SSD (bad)
sofware protection: Samsung (mediocre)
hardware protection: Crucial (best)


Always a 'good idea' but not that relevant to SSD power-loss protection as you need that anyway, even if you have an UPS.

So you're saying a Samsung 850 EVO is low grade? I guess only the Pro is good...

All these other brands are garbage? Maybe they have some sort of protection that is inherent to their design that you're referring to, but if they do, most aren't advertising it, so it's a sort list.

I could really do without the attitude and have no expectation or even desire for you to reply back to me at this point. I only bothered to reply to clarify my questions and for the benefit of the thread should anyone else care to further the discussion and help me out. I won't tolerate being spoken down to, so would just rather you didn't reply at this point.

 

[CiPHER]

As you wish. Sorry i offended you.

 

[aman74]

As you wish. Sorry i offended you.

I'm sorry if I was perhaps overly sensitive, it's possible as I have not slept and I'm trying to gather this info for a Hackintosh build as my current computer died and I'm loaning one, but it has to be returned soon. I'm in a time crunch. I'm also behind on the tech. Last build I did was over 10 years ago, so I'm trying to cram in catching up and learning about how to do a proper Hackintosh build. On top of that there's a good deal on the 850 Evo and I wanted to get enough info to make a decision.

Perhaps you didn't mean it to sound as it did, but there were several comments that come across less than pleasant. I am by no means an imbecile, but some of the technical info is beyond me in this context.

I feel I did a pretty good job of being clear and concise with my questions, despite my lack of knowledge on the subject.

 

[Cerb]

???....not sure what the question is? They are mentioned in the article I linked to.

A search in the page for "MCL" or "PLC" yields no results.

Is are the Crucial drives basically the new Intel drives then? I remember in the early days of SSD the Intel drives were considered the safest. I haven't seen them marketed lately, so wondering if it's now just the Crucial brand that is licensing and marketing the technology? Crucial is Micron and is not just a rebrander like OCZ or Corsair correct? They make their own RAM and SSD's?

No. Intel and Crucial share R&D and manufacturing for memory. Micron can benefit from all the fab R&D Intel does, and Intel can benefit from the memory R&D on memory from Micron (and cheaper NAND chips, but I suspect that they aren't all that much cheaper, compared to buying from a 3rd party, once all is accounted for), along with I'm sure a large host of shared patents. Intel drives with Intel controllers are very good, but Intel decided to ditch them for their consumer drives (the 730 being an outlier). Intel and Crucial SSDs are completely different, even though they may use the same NAND.

 

[Emulex]

Remember there is a difference between Tantalum (SM843T) and Super capacitors in the amount of time they can retain power. Most enterprise SAS drives use super capacitors, where as the SM843T is just an overprovisioned Samsung 840 PRO with 10% OP baked in and tantalum capacitors - which may not provide power fail safe (or may) but quality in the capacitors does count!

 

[Cerb]

Intel's recent ones just use plain electrolytics (I'm sure in part so the ESR and ESL prevents them from draining too fast).

The 320 and 710 didn't use anything exotic, either.

Super capacitors have their niches, but SSDs write so fast that they don't need too much energy, and supercaps need to be managed like batteries, whereas good quality regular caps last basically forever, if specified well (FI, an SSD that shouldn't even run at 70C using 1000hr@105C caps).

 

[billyb0b]

bottom line.... IMO.... it's worth the cost to invest in a quality UPS that will protect from black/brown outs and overloading. and provide features to properly shut down a machine should power go screwy

 

[CiPHER]

I'm sorry if I was perhaps overly sensitive

I guess we just misunderstood each other. Would you object to me sending you a private message about this? I would love to explain some things and hope we can find common ground.

On top of that there's a good deal on the 850 Evo and I wanted to get enough info to make a decision.

Do you need any more information? I guess the conclusion is that 850 EVO is overpriced and perhaps overrated, but still an excellent high-grade SSD for desktop usage. As long as you do not RAID them, i see no reason why not to pick the 850 EVO if you can get a good deal on that and you like the SSD.

Yes there is always some emotional considerations in computer purchase, denying it is futile. And it's perfectly good to not just decide on rational grounds. Why do you think al those techies run overrated machines - they love the technology. Myself included.

But if you were to ask me, i would still prefer the MX100 or MX200 over any Samsung. The price difference is quite big, 20% here in the EU. And the difference in performance is minimal - especially versus the 512GB version of the MX100.

Super capacitors have their niches, but SSDs write so fast that they don't need too much energy, and supercaps need to be managed like batteries, whereas good quality regular caps last basically forever, if specified well.

Indeed. What i heard was that when OCZ talked with clients about their upcoming 'enterprise' PCIe SSD, the clients asked OCZ to use parallel capacitors instead of one big supercapacitor, because of the single point of failure. Whereas an array of capacitors in parallel would be overspecced so that a few of them can fail without compromising its function.

bottom line.... IMO.... it's worth the cost to invest in a quality UPS

If this concerns SSDs, then I can fix this illusion in a matter of seconds:

Open CrystalDiskInfo, select your SSD, look for Unexpected-Power-Loss, look at the raw data (it is in hex; use the menu to set it decimal or convert hex to decimal).

Now ask yourself: have you ever had so many power failures as the (high) number suggest? So how do you account for it? Would a UPS have protected against all those? That is why you need capacitors, or other equivalent form of protection.

 

[aman74]

A search in the page for "MCL" or "PLC" yields no results.

Sorry....MLC and SLC...I was running on no sleep at the time.

No. Intel and Crucial share R&D and manufacturing for memory. Micron can benefit from all the fab R&D Intel does, and Intel can benefit from the memory R&D on memory from Micron (and cheaper NAND chips, but I suspect that they aren't all that much cheaper, compared to buying from a 3rd party, once all is accounted for), along with I'm sure a large host of shared patents. Intel drives with Intel controllers are very good, but Intel decided to ditch them for their consumer drives (the 730 being an outlier). Intel and Crucial SSDs are completely different, even though they may use the same NAND.

Thanks for the info. I had started seeing reference to how Crucial and Micron are one in the same and then I started to think I imagined or "misremembered" the Intel connection. Glad to know I wasn't hallucinating.

I guess we just misunderstood each other. Would you object to me sending you a private message about this? I would love to explain some things and hope we can find common ground.

Sure, that would be appreciated. I have a couple questions for you. I might not be able to get around to communicating too much until tomorrow or so.

Do you need any more information? I guess the conclusion is that 850 EVO is overpriced and perhaps overrated, but still an excellent high-grade SSD for desktop usage. As long as you do not RAID them, i see no reason why not to pick the 850 EVO if you can get a good deal on that and you like the SSD.

Well, at the moment I don't feel confident enough to jump on the sale anyhow. I'm sure there will be others, so that's ok.

In looking at reviews there were a lot of complaints about the price premium not being quite worth it, but it looks like the price has settled and with that sale it was actually less than the Crucial, hence my excitement as it's highly regarded, but then I started to learn about some of these other considerations besides speed.

I know you were saying in real use it doesn't matter a lot and I believe it, but my use is photoshop and audio production and not gaming, so this might be one of the areas where it does matter, I don't know for sure. I wouldn't say I need power user levels and won't be doing 3D rendering and not a lot of video and certainly not 4K for a long time.

If this concerns SSDs, then I can fix this illusion in a matter of seconds:

Open CrystalDiskInfo, select your SSD, look for Unexpected-Power-Loss, look at the raw data (it is in hex; use the menu to set it decimal or convert hex to decimal).

Now ask yourself: have you ever had so many power failures as the (high) number suggest? So how do you account for it? Would a UPS have protected against all those? That is why you need capacitors, or other equivalent form of protection.

What causes that? I think that is one of the reasons I misunderstood. Is this just from PSU ripple or something?

Thanks

 

[CiPHER]

Sure, that would be appreciated. I have a couple questions for you. I might not be able to get around to communicating too much until tomorrow or so.

Will do. Thanks!

Well, at the moment I don't feel confident enough to jump on the sale anyhow. I'm sure there will be others, so that's ok.

In looking at reviews there were a lot of complaints about the price premium not being quite worth it, but it looks like the price has settled and with that sale it was actually less than the Crucial, hence my excitement as it's highly regarded, but then I started to learn about some of these other considerations besides speed.

I know you were saying in real use it doesn't matter a lot and I believe it, but my use is photoshop and audio production and not gaming, so this might be one of the areas where it does matter, I don't know for sure. I wouldn't say I need power user levels and won't be doing 3D rendering and not a lot of video and certainly not 4K for a long time.

Either get the 512GB MX100, or any of the Samsung EVO 850's. Just choose your pick.

What causes that? I think that is one of the reasons I misunderstood. Is this just from PSU ripple or something?

I have not seen any study about this. But i do know that the SSD accounts any power loss as unexpected, unless a STANDBY IMMEDIATE command was issues AND completed before losing power.

So this means, if you hold the power button for 3 seconds to force your computer off, that accounts for unexpected power-loss, since no STANDBY IMMEDIATE command would be sent at all. The same is when you are in BIOS-phase or some other non-modern-OS-phase and you hold the power button just once to turn the pc off, that also counts.

I highly suspect some systems also have a very short power-cycle when you hit the reset button, though on two other systems i could not reproduce this.

My personal conclusion was that Unexpected Power-Loss is inevitable in practical usage. So the SSD needs to have protection against this. But there are more ways to Rome they say, and Samsung has devised its own 'software' implementation to protect against this, and generally works very well and saves the cost for elaborate capacitor solution. But still the NAND could be damaged so that it works but have degraded retention and thus turning into a bad block. As long as the SSD has parity bitcorrection (RAID3/RAID5) then it should be fine but i would prefer a hardware solution instead.

Because aside from the physical damage issue, where some papers write about, i also have concerns about Samsung's software protection potentially having issues with RAID-setups or being part of multi-disk ZFS pools, etc.

I have not seen any documentation about this, but i've seen plenty of situations where major flaws exist in technology that go unknown for the major public for quite some quite if not forever. So it would not surprise me if Samsung SSDs have some potential issues with RAID-setups and other elaborate storage where the disk is part of something bigger and that reverting it would cause an inconsistency with the other parts of the bigger whole.

I should note that this is just theory-crafting from my side. But it has served me well in the past.

 

[Emulex]

Keep in mind servers can hold power up to 2 seconds with a 0 volt drop in power! And remember that hard drives have had the same problem with up to 128meg of ram cache.

In theory a raid controller could send a standby immediate upon detection of voltage error.

Otherwise imagine all those folks who have lost power on consumer SSD - they would be toast with NTFS and no power loss protection (or dead hangs!)

How often do you find you've gotten a corrupted SSD from an abnormal power event on your pc? I've never had this happen!

 

[CiPHER]

Keep in mind servers can hold power up to 2 seconds with a 0 volt drop in power! And remember that hard drives have had the same problem with up to 128meg of ram cache.

Harddrives do not have the same problem because a harddrive will never lose data due to a power failure beyond the last FLUSH command.

All 2nd generation filesystems, including NTFS, have protection against losing recent writes. It uses boundaries where sync metadata writes force the in-flight DRAM of the harddrive to be written to disk before the FLUSH command is completed. This causes the harddrive to lose data only up until the last FLUSH command. When this happens the journal will make sure the filesystem returns to a consistent state, where the metadata matches the stored data.

But with Samsung SSDs this will not work, because Samsung ignores the FLUSH command and will rollback the FTL and thus the visible data to a state BEFORE the last FLUSH-command. This is something filesystems were never designed for. It also defies the whole point of synchronous writes.

It will not matter as long as you revert the whole filesystem and not just a part of it, like when using Samsung SSDs in a RAID-configuration. The failure mode requires at least one Samsung SSD to revert while the others don't. Normal usage will work just fine.

How often do you find you've gotten a corrupted SSD from an abnormal power event on your pc? I've never had this happen!

In the past about 90% of all SSD failures were thanks for FTL corruption; the RAID5 bitcorrection, power-safe capacitors and FTL journaling did mostly fix all these vulnerabilities and made SSDs much more reliable than back in the early days. The FTL and its complexity for various failure modes pretty much is the cause for the high number of SSD failures in the past.