In-Network Round-Trip Time Estimation for TCP Flows


Transmission Control Protocol (TCP)‘s tendency to fill up buffers in the network results in long standing queues. Active Queue Management (AQM) tries to solve this issue but typically requires manual tuning as the optimal parameters depend on the Round-Trip Time (RTT) of the flow, which is unknown to the AQM. In this paper, we use network simulation and supervised learning to train a neural network to infer the RTT of a TCP flow from its queuing behavior. We transfer our model into a real-time network emulator and show that it is able to estimate the RTT with an error of only a few milliseconds.

2nd Workshop “Machine Learning & Networking” (MaLeNe)
Daniel Stolpmann
Doktorand & MLE-Webmaster

Machine Learning for Communication Networks, Low Latency & High Throughput Communication, Mobile Edge Computing, Satellite Communication, Network Coding

Andreas Timm-Giel
Professor & Präsident

Communication Networks, Performance Evaluation, Modelling, Machine Learning for Communication Networks