Summary:

Experiments with 3k nodes are not very reliable. It’s just too much for the lab to handle.
- We are considering ways to run experiments in a more stable and less time consuming way.
Using floodsub when forwarding requests instead of only forwarding to gossipsub peers.
- In Experiment 2, a lightpush client was pushing to 165 nodes each message.
- Should we investigate the impact of this?
Discv5 is still consuming a lot of bandwidth
- Discv5 usage doesn’t scale down for smaller networks. The usage is still high for 200 nodes.
Message delays are longer than expected. We are investigating this issue.
Attempting to grab data from store nodes keeps failing with PEER_DIAL_FAILURE. (Experiment 2)
We are consistently getting these log lines in the filter node servers (Experiment 2):
- reason="waku metatdata request failed: read failed: Stream Closed!"
- reason="waku metatdata request failed: read failed: Stream Reset!"

Experiment 1: Bandwidth