Rapid SSD wear-out ZFS RAID1

[pos-sudo]

Goodevening,

I have a question according to my Proxmox setup, I have a 128GB RAM server with 2X 2TB SSD's in RAID1 (ZFS) configured. I saw that the wear out is around 10% in 34 days, I was researching on the internet and find a possible solution to adjust the ZFS settings like:

recordsize to 16K
logbias=throughput
compression=lz4

It looks like this has a positive effect to the SSD wear out but I see still alot of writes with a ZFS setup is there something I can setup to reduce these high wear out? Because on my server on OVH (they setup Proxmox with MDADM) there is a huge different in wear out while those SSDs has only 450TBW but I have the same amount of VM's (even more) running on it.

Note: My SSDS has a 700TBW

Thanks in advanced.

 

[LnxBil]

Note: My SSDS has a 700TBW

WHAT?????

I assume those are consumer or prosumer SSDs. Don't use them, use Enterprise SSDs!
I'm running Samsung Enterprise SSDs for over 10 years and they're "only" on 110 TBW with ZFS and proxmox for 7 years, before that with mdadm.

 

[pos-sudo]

WHAT?????

I assume those are consumer or prosumer SSDs. Don't use them, use Enterprise SSDs!
I'm running Samsung Enterprise SSDs for over 10 years and they're "only" on 110 TBW with ZFS and proxmox for 7 years, before that with mdadm.

Hello,

Thank you for your reply, haha I was expecting this kind of reply, could you please provide me the SKU of the disk you are using?

 

[LnxBil]

Thank you for your reply, haha I was expecting this kind of reply, could you please provide me the SKU of the disk you are using?

Samsung MZ7WD960 and SM/PM863

 

[pos-sudo]

Samsung MZ7WD960 and SM/PM863

Thanks!

 

[Michael2006]

I had the same issue with ZFS extremely high SSD wearout.
I decided go back to MDADM + LVM + EXT4 configuration. For 1 year of usage I got 5% wearout of the SSD disks with MS SQL databases onboard.
I don't use snapshot functions, so for me there is no reason use ZFS.
Also ZFS is not faster then MDADM + LVM. Periodically ZFS takes too many operations on the disk, so my VMs slows down. I tried many settings, no tangible results.

 

[LnxBil]

there is no reason use ZFS.

Silent data corruption prevention?

 

[pos-sudo]

I had the same issue with ZFS extremely high SSD wearout.
I decided go back to MDADM + LVM + EXT4 configuration. For 1 year of usage I got 5% wearout of the SSD disks with MS SQL databases onboard.
I don't use snapshot functions, so for me there is no reason use ZFS.
Also ZFS is not faster then MDADM + LVM. Periodically ZFS takes too many operations on the disk, so my VMs slows down. I tried many settings, no tangible results.

Dear Michael2006,

Thanks for your reply. Can you provide the disk type you're using, when you had this wear-out issue?

Thanks in advanced!