I’m experiencing an issue in a test scenario with high UDP traffic sent out from containers running on an Atomic Host VM.
The host server has two dedicated 1Gbit NICs: one for ingress traffic and another one for the egress traffic.
When sending small UDP packets (300 bytes) from a container developed by me, I’ve have noticed a reduced throughput and the max used bandwidth is about 159 mbit/sec, while the expected bandwidth is about 600 mbit/sec.
I’ve have repeated the test with iperf docker (https://hub.docker.com/r/mattgruter/iperf/) and the result is the same.
Ingress traffic does not have any kind of issue using both our and iperf docker.
On the ingress NIC I had to set the driver to VirtIO, since with the e1000 we had very bad performances.
While on the egress card I have noticed better performances with e1000 than VirtIO but the throughput is still poor and the used bandwidth is still 159 mbit/sec.
Any suggestion on how to increase egress traffic?