hello!
where should qdiscs be attached?
Inside the LXCs/VMs or on the virtual interfaces on the proxmox hosts?
Why do veth interfaces use the noqueue by default and tap interface use the configured default qdisc?
On my multi queue tap interface the qdisc behavior is even more strange.
It doesn't consistently apply the configured default qdisc to the queues.
From my testing it appears the qdisc should be attached inside the LXCs/VMs?
Otherwise they don't appear to be working correctly, e.g. cake bandwidth limiter is not working.
So it should be noqeue on veth/tap interfaces and configure the qdisc inside the LXCs/VMs?
What about multi queue tap interfaces then?
Just use noqueue as root qdisc? or use mq + noqueue for queues ? (is this even possible?)
And why is it that on Ubuntu LXCs cakes target is limited to 5ms ?
//edit
libvrt v6.9.0 (2020-11-02) changelog:
but that doesn't seem to be the case.
where should qdiscs be attached?
Inside the LXCs/VMs or on the virtual interfaces on the proxmox hosts?
Why do veth interfaces use the noqueue by default and tap interface use the configured default qdisc?
On my multi queue tap interface the qdisc behavior is even more strange.
It doesn't consistently apply the configured default qdisc to the queues.
From my testing it appears the qdisc should be attached inside the LXCs/VMs?
Otherwise they don't appear to be working correctly, e.g. cake bandwidth limiter is not working.
So it should be noqeue on veth/tap interfaces and configure the qdisc inside the LXCs/VMs?
What about multi queue tap interfaces then?
Just use noqueue as root qdisc? or use mq + noqueue for queues ? (is this even possible?)
And why is it that on Ubuntu LXCs cakes target is limited to 5ms ?
//edit
libvrt v6.9.0 (2020-11-02) changelog:
- qemu: Set noqueue qdisc for TAP devices
Set noqueue instead of the former pfifo_fast queue disciplinefor TAP devices. It will avoid needless cost of host CPU cycles andthus improve performance.
but that doesn't seem to be the case.
Last edited: