Open Stack

Folks,

If I am running dpdk in my compute nodes then sure it will reduce
kernel overhead but how does it help to improve guest network
performance?

Inside guest vm it's going to use kernel to copy packet to userspace.
in my example we run high performance haproxy which handles 300k to
500k connections but I have noticed performance is poor and lots of
packet drops.

If I run the same workload on SRIOV vms it works fine and I notice
less kernel context switching inside the guest VM.

I think when you run DPDK in compute then it only help compute but not
your guest network performance.

How do I improve guest performance also with dpdk based compute?