Waku v2 benchmarking: load testing

[jm-clius]

Background

The original plan for Waku v2 suggested scalability and performance at a level necessary for a large, p2p, generalised messaging network. Given that Waku v2 is aimed at resource restricted environments, we are specifically interested in its scalability and resource usage characteristics. The actual scalability limits of a Waku v2 network, however, has never been tested in a realistic production environment.

Proposal

Perform load testing using a test network to understand:

• at what message load performance and reliability starts to degrade
• at what message load performance and reliability drops below a degraded but still acceptable threshold
• what the impact of a larger network and more service nodes is on the above
• if different Waku clients (go/nim/js) exhibit exactly similar or different performance characteristics
• what the effect is on node resource consumption (especially for service nodes)

Note that a third party outside of Status could be contracted to help with these stress tests.

[D4nte]

Can we please add some peer connection metrics:

• How many peers can a node be connected to before reliability starts to degrade
• How many peers can a peer have in its relay mesh before reliability starts to degrade

For both scenarios and to make it relevant to js-waku:

• prefer incoming connections
• on websocket port

Also, I think it may make sense to use digitalocean's droplet as a baseline re resources. (1 vCPU, 2 GB RAM). This would give an idea to operators:

• Capacity of a node running on a low cost cloud service ($5/mo)
• Expected reliability on a raspberry pi (which is actually more powerful than a droplet)

[jm-clius]

Good suggestions, thanks!