The same task, replayed under the sequential, star and full-mesh topologies at the same time. The three panels share one clock; the timeline below marks every LLM call the moment it is sent, so the difference in burstiness shows at a glance.
LLM-call arrivals, shared timeline
One tick per call, at the moment it is sent. The filled area below each row is the number of calls in flight at once, on a common scale.