M2.PCIe-SSDs (Size 2280) without PLP for ZFS possible?

saveZFS

Explorer
Joined
Jan 6, 2022
Messages
87
Hallo,
I want to passtrough two M2-PCIe-SDDs to my Truenas-VM. I want to use it as storage for my ESXi-VMs in a ZFS-Mirror.
My board only supports the form factor 2280. So there is no SSD with power-loss-protection available.
Moreover I can't use a riser card, becase all my PCIe slot are full, yet.
So is it in your opinion posible to use SSDs without PLP for ZFS or will this end in a desaster?
Or is here anybody using a SSD without PLP for ZFS without any problems?
 

Patrick M. Hausen

Hall of Famer
Joined
Nov 25, 2013
Messages
7,776
The strict PLP requirement is for SLOG devices only. You can use your SSDs as regular storage.
 

saveZFS

Explorer
Joined
Jan 6, 2022
Messages
87

Patrick M. Hausen

Hall of Famer
Joined
Nov 25, 2013
Messages
7,776
I run Samsung 960 Pro and 970 EVO Plus. All modern SSDs are "fast" for my needs. The number concerning me most is the TBW. So I recommend you go check that for the devices you are considering. I would not go for less than 600 TBW per TB of storage capacity.
 

Nick2253

Wizard
Joined
Apr 21, 2014
Messages
1,633
To fully understand the SLOG/PLP/SSD dynamic, you need a little background.

Prior to 2015ish, SSDs had a tendency to lie about properly honoring a cache flush. (This made the SSD appear speedier in benchmarks.) If an SSD had PLP, that was a good indicator that the SSD *was* telling the truth about cache flushes, and that become an important "buzz word" for finding a safe SLOG SSD. Now, however, thanks to continuous improvements in flash memory and much wider adoption of SSDs, it's rare to find an SSD that does not honor a cache flush, so this worry is mostly gone. However, just because an SSD now tells you the truth, does not mean that you want it for a SLOG without PLP.

You see, writing data to non-volatile memory, even one made of the worlds fastest flash memory, is incredibly slow compared to volatile memory. The advantage of power loss protection is that the SSD can have a little bit of volatile memory (like DRAM), protected with PLP, in order to quickly suck in data (and honestly report that this data is "safe") before committing to the "slow" task of writing it out to non-volatile memory. Therefore, PLP makes a "synchronous" write (aka, a write that requires confirmation that it was successful) much faster.

Ok, so that brings us to the SLOG itself. The only advantage that a SLOG brings is that synchronous writes can be made faster. The data gets sent to the SLOG, reports that it's been properly and securely written, and then the system can continue with its business. If the SLOG is approximately as fast as the pool, then the SLOG provides no benefits over the pool itself. In any event, when data is written to the pool, that data comes directly from RAM, never from the SLOG. The only time that the SLOG is read is when there is an actual power loss event, so a SLOG's write speed is the only metric that is important. And this brings us back to PLP: when an SSD has PLP, it generally means that it has some kind of fast protected cache that makes its writes much, much faster (and that much more suitable as a SLOG).
 

Arwen

MVP
Joined
May 17, 2014
Messages
3,611
I second @Nick2253 explanation below. A lot of people miss that a SLOG is almost exclusively write only.
In any event, when data is written to the pool, that data comes directly from RAM, never from the SLOG. The only time that the SLOG is read is when there is an actual power loss event, so a SLOG's write speed is the only metric that is important.
 

HoneyBadger

actually does care
Administrator
Moderator
iXsystems
Joined
Feb 6, 2014
Messages
5,112
Prior to 2015ish, SSDs had a tendency to lie about properly honoring a cache flush.
It might be a little Offensive to Call out a specific manufacturer here. Zebras are striped, in an unrelated note.
 

saveZFS

Explorer
Joined
Jan 6, 2022
Messages
87
Thank you very much, then I will search for two cheap SSDs with at least 600 TB TBW. :D
The SSDs with PLP are really very expansive! :(
 
Top