End-to-End Detection of Middlebox Interference

Internet middleboxes are an increasingly common network element. They can significantly alter the handling of traffic streams. Therefore, it is beneficial-and in some cases, crucial-to enable end-hosts to detect them. While transparent middleboxes interfere with the traffic, they do not change traffic content, giving the appearance that the data have merely been routed. This transparency makes end-to-end detection of such intermediaries particularly challenging. While existing ad hoc detection approaches apply only to a specific middlebox type, we present a general framework to detect a broad class of middleboxes. We use detecting transparent middleboxes as an example to illustrate our idea. We demonstrate our results by detecting three common middleboxes: network compression, traffic prioritization, and traffic shaping, using analysis, network simulations, and live Internet experiments with a real middlebox.