Basically, the writes on my only raid 1 on Proxmox is creeping up about a couple 100GBs a day with no explanation. HA and other services people usually recommend to turn off to prevent disk wear are off. I check daily (and weekly) maximums of disk IO on all VMs and it doesn't come close to account the writes. Not sure where it's coming from but wearout on both drives are at 13-14% and was at like 10% a couple of days ago, and same before then. I am using Crucial NVMe drives.
NVMe wearout slowly going up
- Thread starter kolt
- Start date
don't use consumer ssd. Proxmox is doing small sync writes all the times (for cluster, for stats,..).
if you don't have a ssd with power loss protection / supercapacitor, a small 4k write will be amplified to the size of your ssd nand cell (can be multiple MB). (with a supercapacitor protection, the small write are writen in a buffer safely, then flushing to a big write to the ssd nand ssd)
if you don't have a ssd with power loss protection / supercapacitor, a small 4k write will be amplified to the size of your ssd nand cell (can be multiple MB). (with a supercapacitor protection, the small write are writen in a buffer safely, then flushing to a big write to the ssd nand ssd)
You don't say if you use ZFS or LVM mirrors but with ZFS the effect @spirit mentioned is more exaggerated because ZFS is copy-on-write. That means that a small write may cause a large amount of already-written data to be copied. This is good for data integrity but bad for wear.
With consumer hardware you're better off using mdraid + lvm, but you have to install via Debian to get that. I'm at 10% wear-out after 4 years on Samsung 870 EVO.
With consumer hardware you're better off using mdraid + lvm, but you have to install via Debian to get that. I'm at 10% wear-out after 4 years on Samsung 870 EVO.
I see people using Samsung NVMes without issues. I am on ZFS. Is your setup ZFS with Samsung? Might consider replacing both crucials with Samsung nvmes