<i>pHeavy</i>: Predicting Heavy Flows in the Programmable Data Plane

Since heavy flows account for a significant fraction of network traffic, being able to predict has benefited many management applications mitigating link congestion, scheduling capacity, exposing attacks and so on. Existing machine learning based predictors are largely implemented on the control plane Software Defined Networking (SDN) paradigm. As result, frequent communication between data planes can cause unnecessary overhead additional delay in decision making. In this paper, we present pHeavy , scheme predicting directly programmable plane, thus eliminating latency SDN controller. Considering scarce memory limited computation capability includes packet processing pipeline which deploys pre-trained tree models in-network prediction. We have both bmv2 software switch P4 hardware (i.e., Barefoot Tofino). Evaluation results demonstrate that achieved 85% 98% accuracy after receiving first 5 20 packets flow respectively, while reduce size by 5.4x average. More importantly, at line rate switch.