Robust and Listening-Efficient Contention Resolution
share
This paper shows how to achieve contention resolution on a shared communication channel using only a small number of channel accesses – both for listening and sending – and the resulting algorithm is resistant to adversarial noise. The shared channel operates over a sequence of synchronized time slots, and in any slot agents may attempt to broadcast a packet. An agent's broadcast succeeds if no other agent broadcasts during that slot. If two or more agents broadcast in the same slot, then the broadcasts collide and both broadcasts fail. An agent listening on the channel during a slot receives ternary feedback, learning whether that slot had silence, a successful broadcast, or a collision. Agents are (adversarially) injected into the system over time. The goal is to coordinate the agents so that each is able to successfully broadcast its packet. A contention-resolution protocol is measured both in terms of its throughput and the number of slots during which an agent broadcasts or listens. Most prior work assumes that listening is free and only tries to minimize the number of broadcasts. This paper answers two foundational questions. First, is constant throughput achievable when using polylogarithmic channel accesses per agent, both for listening and broadcasting? Second, is constant throughput still achievable when an adversary jams some slots by broadcasting noise in them? Specifically, for N packets arriving over time and J jammed slots, we give an algorithm that with high probability in N+J guarantees Θ(1) throughput and achieves on average O((N+J)) channel accesses against an adaptive adversary. We also have per-agent high-probability guarantees on the number of channel accesses – either O((N+J)) or O((J+1) (N)), depending on how quickly the adversary can react to what is being broadcast.
READ FULL TEXT