This replicates two processes running on the same host, connecting to each other via TCP.

nohup iperf -s -p 5002 >/dev/null 2>&1 &

iperf -c localhost -p 5002

This gives you an rough baseline of what the machine is capable of processing when all processes are on the same machine.

This replicates a process on the host and one inside a container connecting to each other via TCP.

docker run -d -p 5001:5001 --name mapped-iperf benhall/iperf iperf -s

iperf -c localhost

Why the difference performance results?

Create a link between the two containers for them to communicate via friendly names.

docker run -d --name link-server benhall/iperf iperf -s

docker run -it --link link-server:server benhall/iperf iperf -c server

Is the performance different to port mapping in the previous test?

This uses a Docker Network to create a bridge. Containers communicate over that bridge.

docker network create iperf

docker run -d --net=iperf --name network-server benhall/iperf iperf -s

docker run -it --net=iperf benhall/iperf iperf -c network-server

Any difference to links? Run the test a few times and see.

Containers using a shared network.

docker run -d --name shared-server benhall/iperf iperf -s

docker run -it --net=container:shared-server benhall/iperf iperf -c localhost

What is happening underneath? How does the performance look?

This takes a look at our original test, but running as a container.

Client inside a container, connecting to host via the Host Network

docker run -it --net=host benhall/iperf iperf -c localhost -p 5002

Server inside a container, connecting to client on host via the Host Network

First, remove the container which is mapped to port 5001 docker rm -f mapped-iperf

docker run -d --net=host benhall/iperf iperf -s

iperf -c localhost

Wow?