TBSW: Time-Based Sliding Window Algorithm for Network Traffic Measurement

Network traffic measurement is vital to network applications such as network anomalies detection, capacity planning, and congestion alleviation. Traditional switches sample a small portion of packets and send them to CPU for analysis. However, with sampling rates as low as 0.1%, sampling methods are inaccurate. Faced with the predicament, emerging programmable switches achieve measure functionality to a certain extent. Nevertheless, existing methods depend on CPU to drain and clear the records periodically, which brings in too much communication latency. To avoid disadvantages above and explore the processing ability of modern programmable switches, we proposed a time-based measuring algorithm named TBSW. TBSW is designed to measure the number of packets during an interval accurately with little hardware resource usage. To the best of our knowledge, TBSW is the first to utilize the time information for network traffic measurement. We prototype TBSW in P4 language and run it on Barefoot Tofino switch. Results showed TBSW achieves high efficiency and moderate accuracy at line speed.