Hey guys,
So I'm migrating several existing kvm qcow images to proxmox (love the web interface) by creating new KVM's in proxmox and replacing their qcow files.
On several KVM instances (not all) I started to experience VERY high ping times to other machines, including those within the same bridge and same subnet. Ping times went from sub 1ms to over 400-500ms, oddly this seemed to effect other operations on the machine, even via VNC view.
After much playing/reconfiguring (2 days of fiddling with settings/rebuilding machines), I noted the following:
1) the high ms ping times and subsequent issues happened only on VMs that had heavy network IO (e.g. NFS mounts).
2) Once the issue started, no amount of daemon/network restarting helped.
3) It appeared that the rx/tx buffers were so full that the VMs started to experience 30%-80% wait io times in top without individual processes using the cpu. Load often reached over 2.00 on machines with 2 virt processors.
4) Without any other mods, issues resolved immediately upon switching from virtio to e1000 network interfaces.
Most of my VMs were running kernel version 2.6.28-15-server with Ubuntu 9.04 or 9.10. Many different machines had wide varying configs and purposes (e.g. mail server, ftp server, dlna music server, etc).
I will be trying to duplicate such issues under different VM kernels, but thought I would start the thread now to see if others had similar experience.
Only individual VMs were effected, as other running VMs had normal ping times/network performance during the same time period. My concern is that this may be related to the host PVE kernel version....
Thanks,
David
So I'm migrating several existing kvm qcow images to proxmox (love the web interface) by creating new KVM's in proxmox and replacing their qcow files.
On several KVM instances (not all) I started to experience VERY high ping times to other machines, including those within the same bridge and same subnet. Ping times went from sub 1ms to over 400-500ms, oddly this seemed to effect other operations on the machine, even via VNC view.
After much playing/reconfiguring (2 days of fiddling with settings/rebuilding machines), I noted the following:
1) the high ms ping times and subsequent issues happened only on VMs that had heavy network IO (e.g. NFS mounts).
2) Once the issue started, no amount of daemon/network restarting helped.
3) It appeared that the rx/tx buffers were so full that the VMs started to experience 30%-80% wait io times in top without individual processes using the cpu. Load often reached over 2.00 on machines with 2 virt processors.
4) Without any other mods, issues resolved immediately upon switching from virtio to e1000 network interfaces.
Most of my VMs were running kernel version 2.6.28-15-server with Ubuntu 9.04 or 9.10. Many different machines had wide varying configs and purposes (e.g. mail server, ftp server, dlna music server, etc).
I will be trying to duplicate such issues under different VM kernels, but thought I would start the thread now to see if others had similar experience.
Only individual VMs were effected, as other running VMs had normal ping times/network performance during the same time period. My concern is that this may be related to the host PVE kernel version....
Thanks,
David
