Motivation
I'm running into unclear performance characteristics as the number of agents increases in the planning loop. I want to benchmark and track latency as agent counts scale, so I can spot bottlenecks and optimize accordingly.
Proposed approach
- Add a benchmarking script that spins up planning loops with variable agent counts (e.g., 1, 5, 10...)
- Measure and log end-to-end latency for each run, ideally with breakdowns (planning, execution, communication time)
- Store results in a simple CSV or JSON format for comparison over time
Motivation
I'm running into unclear performance characteristics as the number of agents increases in the planning loop. I want to benchmark and track latency as agent counts scale, so I can spot bottlenecks and optimize accordingly.
Proposed approach