How the experiment is done

Constant experiment parameters

Experiments description:

Notes:

IDONTWANT was introduced on nim-libp2p 1.2. We can see the effect of IDONTWANT control messages in the 50KB and 500KB plots, where the bandwidth from 1.2 to onwards gets decreased with respect to 1.1.

Bandwidth incrase in 1.6 is due to a Kubernetes worker misconfiguration.

Current issues:

We are currently working on stability issues in DST lab. Since version v1.11.0, it is hard to get stable bandwidth metrics, so what we are trying to observe is that there are not mayor differences with pervious versions.

Still, we can do the following observations:

• There are no major regressions in terms of bandwidth usage with respect to previous versions
• Quic is still unstable
  • GC memory for 50KB increases non stop
    • The delays vary between 20ms to 2 seconds, I don't know if it is related to peers getting disconnected (no artificial delay here).,
    • Metrics are weird as in previous versions
    • Looks like the mesh cannot reach a stable state
      • We check healthiness by getting how many peers have between d_low and d_high peers. Something that is reached almost instantly in Yamux and Mplex, looks like cannot be achieved in Quic.

Nim-libp2p bandwidth plots:

100 bytes payload:

1 KB payload: