JetStream: Achieving Predictable Gossip Dissemination by Leveraging Social Networks Principles

Gossip protocols provide probabilistic reliability and scalability, but their inherent randomness may lead to high variation in (received) message overheads at different nodes. This paper presents techniques that leverages simple social networks principles to enable nodes to select gossip targets intelligently. These simple heuristics achieve a more uniform message overhead at each node, while at the same time reducing the latency of gossip spread (by up to 25%) and also lowering the system-wide gossip network traffic. We experimentally compare our system, called JetStream, against canonical gossip as well as gossip on the Chord overlay. Intuitively, JetStream seeks to make gossip spread more deterministic and predictable, while still inheriting its scale and reliability. JetStream also provides an added benefit by reducing overall network bandwidth utilization if there is a low amount of sustained gossip injection.