Simple and affordable network traffic classification

What is it about?

Network traffic is heterogeneous, and streams of data are often categorized into long or small flows. Knowing such class of a flow in advance is a valuable insights for various networking tasks, including scheduling and monitoring. In this paper we propose a system named DUMBO that predicts if a flow will be long (elephant) or small (mouse). This system integrates a lightweight traffic classifier whose predictions are used to enhance three tasks: packet scheduling, inter-arrival times distribution estimation and flow length estimation.

Featured Image

Photo by Bisakha Datta on Unsplash

Photo by Bisakha Datta on Unsplash

Why is it important?

Providing early knowledge on the size of traffic flows is a crucial but complex endeavor that can enhance many networking tasks, ranging from scheduling to congestion control. In this paper, we take a first step toward this goal, by providing coarse-grained hints on flow size in the form of a binary classification between elephants and mice flows.

Perspectives

“

We propose an affordable, yet effective, Machine Learning networked system that can enhance many downstream tasks from simple predictions on flow length. Code is open and our simulator can be used with any real traffic traces.

”