I am impressed every day by the quality of proxmox, whose main asset is its stability. thank you for this work
currently on proxmox 5.4 we are not able to reach 10G and +
The main reason is the linux kernel network stack design whose implies high context switches that affect performance
The nic offload and the TSO/GRO optimizations bring a bandwidth gain but nothing at the level of the PPS which has a lot of difficulty to exceed the million (according to our use cases)
many alternative projects use network stacks in user space and we know that performance can be x10
but none of them are proxmox compatible
An other approch consist of using kernel fast-path, but the performance gain is not evident...
I would like to know what the development lead thinks about how to perform +10Million PPS on proxmox
have any tests been performed ?
currently on proxmox 5.4 we are not able to reach 10G and +
The main reason is the linux kernel network stack design whose implies high context switches that affect performance
The nic offload and the TSO/GRO optimizations bring a bandwidth gain but nothing at the level of the PPS which has a lot of difficulty to exceed the million (according to our use cases)
many alternative projects use network stacks in user space and we know that performance can be x10
but none of them are proxmox compatible
An other approch consist of using kernel fast-path, but the performance gain is not evident...
I would like to know what the development lead thinks about how to perform +10Million PPS on proxmox
have any tests been performed ?