List of SSDs with Power Loss Protection

[dmitrios]

I tried searching and did not find... so I propose to build a list of all SSDs (PCI-E or M.2 NVME and SATA) that have Power Loss Protection.
Inspired by this and this.

For my build I am going to buy a s/hand HBA from eBay (have only one PCI-E slot), flash it into IT or IR (because why not? gives extra option to have the card do RAID if used outside of FreeNAS), and then have a SATA SSD (with PLP) as boot drive (because why not? lol) and then maybe another SATA SSD of larger capacity to store VMs or be used as a SLOG...

[Work-in-progress] - Do all of these have PLP? https://www.intel.co.uk/content/www...d-state-drives/data-center-ssds/view-all.html

Intel -> Find products with Enhanced Power Loss Data Protection


PCI-E NVME SSDs
Intel DC Pxxxx series

M.2 NVME SSDs


M.2 SATA SSDs
Samsung SM953

2.5" SATA SSDs
Intel DC S3700 [1]
Intel DC S3500 [1]
Intel DC S3610

Samsung PM953
Samsung PM863 ("for read-intensive applications")
Samsung SM863 ("for write-intensive applications")

Micron M600DC
Micron M500DC?

---

[1] Intel's Power Loss Imminent Technology Brief
The Intel® SSD DC S3700 Series and S3500 Series with SATA 3.0 interface contain enabled PLI Technology hardware and the firmware to manage the SSD through the power loss event.


Amazingly, WD claim to have PLP in their "Green" SSDs. Not sure how much this is to be relied upon...

 

[arekm]

2.5" SATA SSD:

Intel S4510 ("Enhanced Power Loss Data Protection")
Micron 5200 series ("Enhanced power-loss data protection with data protection capacitor monitoring")
https://www.micron.com/-/media/clie.../ssd_power_loss_protection_white_paper_lo.pdf

List of common (consumer or enterprise) ssds which do not have the protection would be good, too.
Usually looking at PCB photos and lack of capacitors there mean there is no power protection. WD green PCB photos show no capacitors bank for example.

Edit:
Micron 1100 series ("Power-Loss Protection (data at rest)")
was there but I think PLP for "data at reset" is just marketing bulshit and it doesn't provide any protection on power loss while drive is being used.

 

[HoneyBadger]

Amazingly, WD claim to have PLP in their "Green" SSDs. Not sure how much this is to be relied upon...

I know I'm thread-necroing a little here, but I think this is an important resource to maintain, so here's your explanation.

The WD Green SSD has no RAM buffer - there's physically no nonvolatile cache on the drive itself. By design it writes directly to NAND. To offset the slow performance of the TLC NAND, it treats a portion in a manner similar to SLC and uses it as a "write cache" which is then folded into TLC cells and written again.

This means that while it is safe from power failures, the endurance numbers are incredibly low and it shouldn't be used for an SLOG device, unless you just want to see how quickly you can kill it.

Amusingly enough, the Intel Optane lineup is also "direct to NAND" writes but they use much, much faster raw storage (and a different type of programming entirely, but that's another thread I could write) which means that they have both greater native performance without the "SLC cache cheating" and better endurance.

 

[WaltR]

I tried searching and did not find... so I propose to build a list of all SSDs (PCI-E or M.2 NVME and SATA) that have Power Loss Protection.
Inspired by this and this.

For my build I am going to buy a s/hand HBA from eBay (have only one PCI-E slot), flash it into IT or IR (because why not? gives extra option to have the card do RAID if used outside of FreeNAS), and then have a SATA SSD (with PLP) as boot drive (because why not? lol) and then maybe another SATA SSD of larger capacity to store VMs or be used as a SLOG...

[Work-in-progress] - Do all of these have PLP? https://www.intel.co.uk/content/www...d-state-drives/data-center-ssds/view-all.html

Intel -> Find products with Enhanced Power Loss Data Protection


PCI-E NVME SSDs
Intel DC Pxxxx series

M.2 NVME SSDs


M.2 SATA SSDs
Samsung SM953

2.5" SATA SSDs
Intel DC S3700 [1]
Intel DC S3500 [1]
Intel DC S3610

Samsung PM953
Samsung PM863 ("for read-intensive applications")
Samsung SM863 ("for write-intensive applications")

Micron M600DC
Micron M500DC?

---

[1] Intel's Power Loss Imminent Technology Brief
The Intel® SSD DC S3700 Series and S3500 Series with SATA 3.0 interface contain enabled PLI Technology hardware and the firmware to manage the SSD through the power loss event.


Amazingly, WD claim to have PLP in their "Green" SSDs. Not sure how much this is to be relied upon...

Crucial's MX500 series SATA has Integrated Power Loss Immunity, which is amazing for a $42 250 GB Consumer SSD

 

[HoneyBadger]

Crucial's MX500 series SATA has Integrated Power Loss Immunity, which is amazing for a $42 250 GB Consumer SSD

Hi WaltR. Unfortunately Micron/Crucial's "power loss immunity" is not the same as protection for data in-flight - the implementation on the Crucial MX consumer drives is only guaranteed to protect data at rest from being corrupted by a new write in progress when power is lost. This is not the same as the end-to-end power-loss-protection offered on enterprise drives, which will protect "pending writes" that are in the drive's DRAM cache.

Thankfully the drive does obey the "flush cache" command properly, and will in fact push its DRAM to NAND when asked; but this is limited by the write speed of the NAND and whether or not there are pages free to be "fast programmed" as SLC, so it will likely be slow.

If you do have one, please feel free to benchmark it using the diskinfo command found in the thread in my signature and provide the results. I imagine the SLC write cache could let it deliver some fairly good numbers for bursty workloads, but sustained writes could cause it to choke.

And with all that said - welcome to the forums. :)

 

[otpi]

Maybe the Ironwolf ssd? I believe it has "power loss data protection circuit", but could be it's the same caveat as the cruicial MX500 series.

 

[HoneyBadger]

Maybe the Ironwolf ssd? I believe it has "power loss data protection circuit", but could be it's the same caveat as the cruicial MX500 series.

It appears to be based on the enterprise "Nytro" SSDs, which do have full PLP for their write cache; anyone want to be the guinea pig? ;)

Edit - We have a guinea pig, the results are in (they're bad [1] ) and look what I just discovered in the Ironwolf 110 manual [2]

Power Loss Data Protection  In-process writes to the NAND are completed in the event of an unexpected power loss

So confirmed, this is just the same as the MX500 "in progress NAND writes" and not full end-to-end PLP, the cache is still volatile.

[1] https://www.ixsystems.com/community...nding-the-best-slog.63521/page-11#post-552402
[2] https://www.seagate.com/www-content/manuals/ironwolf-110-ssd/pdf/IronWolf_110_SSD_PM-100841919_B.pdf

 

[Hobbel]

PCI-E NVME SSDs
Samsung 1725b

 

[HoneyBadger]

was there but I think PLP for "data at reset" is just marketing bulcrap and it doesn't provide any protection on power loss while drive is being used.

A bit of both honestly ... the marketing of "protecting your data at rest" is good because it means that the drive won't knowingly corrupt existing good data by failing to completely program a page.

See this rather dated but still relevant PDF from USENIX FAST '13:

https://www.usenix.org/system/files/conference/fast13/fast13-final80.pdf

Drives could have flying/shorn writes, metadata corruption in the FTL, or other nastiness. A limited amount of PLP and/or the necessary firmware safeguards really ought to exist in all drives in order to prevent this kind of thing, it's a great value to consumer workloads.

The issue is that SLOG writes aren't really a "consumer workload" per se - most users are willing to accept a couple KB of lost data. Oh no, I need to redownload that MP3 from iTunes. Oh no, I need to redownload that recipe from Pinterest. Oh no, I need to rewrite that Facebook post. SLOG and other sync write workloads live nicely in the "enterprise workload" category, where once the upstream device believes the write was committed, it damn well needs to be committed, every single 1 and 0 of it. That's where full PLP, including the volatile DRAM cache, comes into play.

Most modern drives are decent enough to be honest about their PLP capabilities once they're actually connected - if they have the necessary juice to protect DRAM, they can consider it as non-volatile. If they don't, most devices will honor the flush request by committing to NAND - this is slow, and which is why the vast majority of SSDs make for honest but poorly performing SLOG devices. (No one wants to have their write speeds measured in KB/s.)

Older devices however are occasionally dishonest and lie about their PLP capabilities. I'm reasonably certain I have one in my possession, it's an old OCZ Vertex 2 on dodgy firmware that's only recognized about 50% of the time it's powered on. According to benchmarks, it's supposedly capable of better SLOG write speeds than one of the best SATA SSDs, the Intel DC S3700. But the Vertex2 doesn't have PLP for the DRAM cache; so it's just lying and claiming it does - when it gets a "flush cache" request, it just says "okay sure I did it" and doesn't actually protect the data.

 

[GrantWilson]

A bit of both honestly ... the marketing of "protecting your data at rest" is good because it means that the drive won't knowingly corrupt existing good data by failing to completely program a page.


See this rather dated but still relevant PDF from USENIX FAST '13:


https://www.usenix.org/system/files/conference/fast13/fast13-final80.pdf


Drives could have flying/shorn writes, metadata corruption in the FTL, or other nastiness. A limited amount of PLP and/or the necessary firmware safeguards really ought to exist in all drives in order to prevent this kind of thing, it's a great value to consumer workloads.


The issue is that SLOG writes aren't really a "consumer workload" per se - most users are willing to accept a couple KB of lost data. Oh no, I need to redownload that MP3 from iTunes. Oh no, I need to redownload that recipe from Pinterest. Oh no, I need to rewrite that Facebook post. SLOG and other sync write workloads live nicely in the "enterprise workload" category, where once the upstream device believes the write was committed, it damn well needs to be committed, every single 1 and 0 of it. That's where full PLP, including the volatile DRAM cache, comes into play.


Most modern drives are decent enough to be honest about their PLP capabilities once they're actually connected - if they have the necessary juice to protect DRAM, they can consider it as non-volatile. If they don't, most devices will honor the flush request by committing to NAND - this is slow, and which is why the vast majority of SSDs make for honest but poorly performing SLOG devices. (No one wants to have their write speeds measured in KB/s.)


Older devices however are occasionally dishonest and lie about their PLP capabilities. I'm reasonably certain I have one in my possession, it's an old OCZ Vertex 2 on dodgy firmware that's only recognized about 50% of the time it's powered on. According to benchmarks, it's supposedly capable of better SLOG write speeds than one of the best SATA SSDs, the Intel DC S3700. But the Vertex2 doesn't have PLP for the DRAM cache; so it's just lying and claiming it does - when it gets a "flush cache" request, it just says "okay sure I did it" and doesn't actually protect the data > https://www.achieveessays.com/write-my-term-paper.html

Hello,
I had endless premature failures using OCZ SSD. Now, as I grew older, I started choosing between Samsung or Intel SSDs. Some years ago I used a dead OCZ Vertex2 SSD built around Sandforce SF-1222, however, I got rid of it after it did'nt get recognized by the BIOS.

 

[sretalla]

Amazingly, WD claim to have PLP in their "Green" SSDs. Not sure how much this is to be relied upon...

Until the TRIM problem is solved, these drives should be avoided... you will get scrub errors and file corruption every time with these unless you disable TRIM for now (not a great thing to do as it's system-wide).

 

[Mike Guilmette]

So would a pair of MX500's work good for the boot device? I'm currently using a single 32gb Supermicro SATADOM, of which I can't locate a brand new one for a decent price any longer. I'd like to have some redundancy for my boot and would prefer to have something with some form of PLP.

Not opposed to getting a pair of the Intel DC S3500's but cost savings of the consumer based drives might be worth the trade off...

 

[Jessep]

I seem to recall there is an issue with the MX500 and FreeNAS. Pending sector issues? A quick search didn't find it.

 

[Mike Guilmette]

Ah ok. I will probably pick up a pair of DC S3500's on ebay then. Thanks!

 

[WaltR]

I'm using MX500's as boot drives no problem. Never hear of any issues. If you want to move up the food chain, I suggest the Micron 5200 series. It's what 45 Drives uses for their FreeNAS builds

 

[sretalla]

Seems Samsung has new devices with PLP... here's one example https://www.samsung.com/us/business/products/computing/ssd/enterprise/983-dct-960gb-mz-1lb960ne/

 

[Jessep]

I'm using MX500's as boot drives no problem

https://www.ixsystems.com/community/threads/what-does-critical-error-currently-unreadable-pending-sectors-mean.64309/

 

[WaltR]

https://www.ixsystems.com/community/threads/what-does-critical-error-currently-unreadable-pending-sectors-mean.64309/

Haven't seen the bug, but then I run FreNAS as a VM on ESXi which owns the boot drives and it doesn't say boo about the bug. FreeNAS only gets a boot volume and full access to the storage disks. If I decide I want to use another ZFS solution later, I just build an new VM and import the pools. No mucking with boot loaders or partition mapping.

You can get really good deals on Micron's 5200 enterprise drives these days. 5200 Eco 480 GB, which are serious overkill for boot drives, are available at CDW of all places for $254.55 a pair delivered. 960's are $149.99 delivered on New Egg.

Market prices on Intel® SSD DC S3610 Series drives with PLP are low. The line is discontinued. but they are solid

 

[Octopuss]

I believe Samsung SM961 has power loss protection as well.

 

[patrickjp93]

I believe Samsung SM961 has power loss protection as well.

Nope, just had one kill my NUC actually. I mean, MAYBE that's Windows 10's fault, but having the drive come up as unrecognized boot media altogether? It re-imaged just fine. No SMART errors reported...