So I posted an issue on here last year regarding high wearout levels on my ZFS raid 1 which hosts both the OS and VMs. This is based off of Crucial P3 NVME ssds which I learned are not great, and something with write amplification is likely the source of the wearout. I have done some quick research and learned that Samsung 980s seem to be working fine with low wearout levels but I understand they don't come with power loss protection. Just wondering if people have a similar setup with Samsung 980 NVMes and if they have similar issues and to see if these drives are good to use.
Replacing nvme ssds
- Thread starter kolt
- Start date
If you want good performance and care about wear you should buy something with power-loss protection. The PLP is the black magic that makes performance and wear not totally horrible once there are sync writes involved.
And yes, those will cost more. But storage isn't different to anything else. If you aren't willing to spend some money you won't get something with high quality that will last long.
And yes, those will cost more. But storage isn't different to anything else. If you aren't willing to spend some money you won't get something with high quality that will last long.
Last edited:
This (PLP vs non-PLP) has been here discussed many times. 
There's people who do not care for it and I think that is fine. I suppose since you had P3 there before, it's not anything serious production.
About 980s, there's a difference between 980 and 980 PRO. Please do not do (another) DRAM-less SSD. It's a horrible use case fit. About 980 Pros, they were known to have firmware issues that caused premature deaths, make sure if you get one of those that you update the firmware if it was old stock.
Make sure you get something with high TBW (consumer ones with higher capacities). And tweak your ZFS or just do not use ZFS at all in these cases. As always, run backups.
And yes, it's 'unsupported', but you don't run a datacentre, do you?
@Dunuin I am now testing Kingston DC600M for the very reason we discussed about PLP vs non-PLP. I virtualised nodes ... on a node ... on ZFS ... and running all that on that. Will report of the PLP really saved any shredding with lots of writes or not at all. I think it's really proprietary and down to each manufacturer, though.
There's people who do not care for it and I think that is fine. I suppose since you had P3 there before, it's not anything serious production.About 980s, there's a difference between 980 and 980 PRO. Please do not do (another) DRAM-less SSD. It's a horrible use case fit. About 980 Pros, they were known to have firmware issues that caused premature deaths, make sure if you get one of those that you update the firmware if it was old stock.
Make sure you get something with high TBW (consumer ones with higher capacities). And tweak your ZFS or just do not use ZFS at all in these cases. As always, run backups.
And yes, it's 'unsupported', but you don't run a datacentre, do you?
@Dunuin I am now testing Kingston DC600M for the very reason we discussed about PLP vs non-PLP. I virtualised nodes ... on a node ... on ZFS ... and running all that on that. Will report of the PLP really saved any shredding with lots of writes or not at all. I think it's really proprietary and down to each manufacturer, though.
Don't forget to run some DBs with good amount of writes for some real world sync writes.
Last edited:
Thanks for advice. I think I am going to get a pair of D3 S4510's which are Intel and has what you mentioned. It's a tough decision to replace the nvme's but after thinking about it more, I don't need the crazy speeds it comes with.
My next question, can I just replace both drives in raid 1 with the the 2 ssds one by one, following "Changing a failed drive" in https://pve.proxmox.com/wiki/ZFS_on_Linux#_zfs_administration ?. Both drives have a 'BIOS boot', 'EFI', and 'ZFS' partitions so I think I can ignore "Changing a failed bootable device"?
Then you have to follow "Changing a failed bootable device" or you won't be able to boot your PVE once you replaced the second disk.