Proxmox / Qemu jemalloc
- Thread starter freebee
- Start date
Hi,
yes, would need to have a good reason for it and would need to be re-evaluated. We actually had enabled it in the past and had to disable it, because of an issue.
yes, would need to have a good reason for it and would need to be re-evaluated. We actually had enabled it in the past and had to disable it, because of an issue.
The pointed problem was solved.
The VM allocates more memory than expected and in some way is unpredictable when we need to partition the ram memory resource.
A suggestion is in apt install have the qemu-jemalloc version. The jemalloc libs can be redistributed without any problems in the license.
I use it on other projects like MySQL and Nginx with success.
The VM allocates more memory than expected and in some way is unpredictable when we need to partition the ram memory resource.
A suggestion is in apt install have the qemu-jemalloc version. The jemalloc libs can be redistributed without any problems in the license.
I use it on other projects like MySQL and Nginx with success.
Yes, but in the commit Stefan only says "it appears to be related" to this issue. And code on our side also changed in the meantime, that's why re-enabling jemalloc would need to be carefully evaluated first.
Did you already take a look at: https://pve.proxmox.com/wiki/Dynamic_Memory_Management ? Maybe something mentioned there helps in your case.
Yes, o look at it.
Some windows virtual machines (big ones like 30-40 cpus) the BALLOON/guest service stop.
Another thing is about some database virtual machines: the recommendation is to turn off the ballon, fixing to a static memory size.
For last, in a reboot, the KSM needs some time to work, and this leaves less CPU time for VMs.
A new approach (for me) is fixing to a static ram memory for all VMS, but I need to optimize the total memory cost to the hardware.
There are also patches for supporting virtio-mem on the pve-devel mailing list. Not sure when these will land though, but that would also give you more control.
But how about the resources ?
I have a vm with 8GB ram. The qemu running this virtual machine is occupying 13.9GB (81 days up). Is almost 73% more.
My impression is fragmentation.
