Adaptive Network Protocol Selection: A Machine-Learning Approach

Abstract

We introduce "optimization through protocol selection" (OPS) as a technique to improve bulk-data transfer on shared wide-area networks (WANs). Instead of tuning the parameters of a network protocol, our empirical results show that the selection of the protocol itself can result in up to four times higher throughput in some key cases. However, OPS for the foreground traffic (e.g., TCP CUBIC, TCP BBR) depends on knowledge about the network protocols used by the background traffic (i.e., other users). Yet, global knowledge can be difficult to obtain in a dynamic distributed system like a WAN. Therefore, we introduce and evaluate a machine-learning (ML) approach for recognizing the background mix of protocols on a shared network. We build and empirically evaluate several ML classifiers, trained on local round-trip time (RTT) time-series data gathered using "passive probing" or "active probing", to recognize the mix of TCP CUBIC versus TCP BBR congestion control algorithms (CCAs) in the background with an accuracy of up to 95%. Then, a decision process selects the best protocol to use for the new foreground transfer, so as to maximize throughput while maintaining fairness (i.e., OPS). Lastly, we describe the design, implementation, and evaluation of iPerfOPS, the first tool that uses OPS to perform bulk-data transfer. The new tool is a substantially modified version of the well-known iPerf tool, and is an end-to-end implementation that incorporates previous research results. Our evaluation of iPerfOPS shows a bandwidth utilization close to the fair (i.e., equal) sharing if used with an appropriate probing pattern.