Leveraging Zipf's law for traffic offloading

摘要: Internet traffic has Zipf-like properties at multiple aggregation levels. These suggest the possibility of offloading most from a complex controller (e.g., software router) to simple forwarder commodity switch), by letting handle very limited set flows; heavy hitters. As volume flows is highly dynamic, maintaining reliable hitters over time challenging. This especially true when we face limit in non-offloaded combination with constraint size hitter or its rate change. We propose selection strategy that takes advantage different scales. Based on real traces, show our able offload while limiting change set, suggesting feasibility alternative router designs.