Scalable TCP Throughput Limitation Monitoring

The analysis of TCP throughput limitations is purposed to determine the limitation preventing a TCP connection to increase its throughput. Monitoring throughput limitations of TCP connections enables detecting network misconfiguration, overload, and other anomalies while connections are still active in the monitored network. Therefore, TCP throughput limitation monitoring enables reactions to such incidents in real-time to improve per-flow and network-level performance. This paper presents a multi-threaded TCP throughput limitation monitoring framework providing scalability due to fully parallelized analysis pipelines. We benchmark our framework's performance and use our implementation to conduct a case study on real-world Internet traffic captured on 10 Gbit/s links. Our framework shows to scale up linearly and to be capable of monitoring workloads of several Gbit/s distributed on several ten thousands of concurrent flows on commodity hardware. We use a real-world Internet trace set including over 9 million TCP flows to assess the share of flows suitable for our monitoring approach, interpret distributions of measured limitation scores, and estimate the throughput limitations of the analyzed TCP connections.