Hi,
I have/had the Proxmox settings for ARCsize changed to 32G min and 64G max as I have 128GB ram on the system. Also, as visible in one of the screenshots, KSM is activated due to a similar VM used. This went all well until today when I noticed a huge performance hit. Then I realized the I/O delay spiking up to ~15%, which never happened in the past. Max was always 4.3-4.5%. Then I realized my whole ram usage was as low as never before with most production VM's running. So I thought the cache must have been gone and checked ARCstats. They showed that ARC collapsed from 62G to 68M. KSM is set to jump in at 35%.

This clearly explains the performance hit, but why does ARC collapse?

The only change in the system made was an update of libnns3 before the incident.
Start-Date: 2020-07-18 12:39:07
Commandline: apt upgrade
Upgrade: libnss3:amd64 (2:3.42.1-1+deb10u2, 2:3.42.1-1+deb10u3)
Another workload done was cloning one single VM between 13:25 and 14:55. That's all.
Any thoughts would be much appreciated. Thank you in advance.
I have/had the Proxmox settings for ARCsize changed to 32G min and 64G max as I have 128GB ram on the system. Also, as visible in one of the screenshots, KSM is activated due to a similar VM used. This went all well until today when I noticed a huge performance hit. Then I realized the I/O delay spiking up to ~15%, which never happened in the past. Max was always 4.3-4.5%. Then I realized my whole ram usage was as low as never before with most production VM's running. So I thought the cache must have been gone and checked ARCstats. They showed that ARC collapsed from 62G to 68M. KSM is set to jump in at 35%.

This clearly explains the performance hit, but why does ARC collapse?

The only change in the system made was an update of libnns3 before the incident.
Start-Date: 2020-07-18 12:39:07
Commandline: apt upgrade
Upgrade: libnss3:amd64 (2:3.42.1-1+deb10u2, 2:3.42.1-1+deb10u3)
Another workload done was cloning one single VM between 13:25 and 14:55. That's all.
Any thoughts would be much appreciated. Thank you in advance.
Last edited: