Detecting Heavy Hitters in Network-Wide Programmable Multi-Pipe Devices

A way to contribute to network management involves detecting high-impact traffic flows, known as Heavy Hitters. Heavy Hitters are flows that account for the majority of bytes transmitted over the network, consequently consuming more resources. The use of programmable hardware, such as switches and DPUs, allows for the detection of these flows in-line rate, aiming to optimize their performance. While the literature reveals an extensive analysis of detection in single-pipe switches, this study presents two approaches to identify Heavy Hitters in programmable switches with multiple pipes. One approach has an accumulator in the switch, which centralizes data from all pipes and communicates with the control plane. In the other approach, communications with the control plane are independent for each pipe. Both approaches were developed and validated through an emulator, demonstrating effectiveness and improvement in detection in multi-pipe switches compared to single-pipe switches.