A Longitudinal Analysis of Traffic Policing Across the Web

Internet traffic has increased 5× in 5 years [4], much of it from high-volume services such as cloud storage and peerto-peer sharing, and from the explosion of streaming video. YouTube and Netflix alone combine to contribute nearly half of the traffic to North American Internet users [8]. This is driven by the fact that these services deliver high-volume traffic, and by the vast popularity of some of these services— YouTube has one billion unique users per month [12] and more than 12% of the US population uses Netflix [6]. This high-volume traffic and its performance is important: users want a quality Internet experience; content providers rely on it for revenue; and Internet Service Providers (ISPs) must cope with delivering its volume. While content providers want to maximize the user quality of experience for their services, an ISP needs to accommodate traffic from a multitude of services and users, often through different service agreements such as tiered data plans. High-volume services like streaming video and bulk downloads (e.g., software updates) that require high goodput must coexist with smaller volume Web services that require low latency. The question of how to manage high-volume traffic has generated both technical and policy discussions. Disputes arise over how to efficiently deliver it [9], and the question of how to handle the growing volumes of traffic has become important enough that even the President of the United States recently weighed in [7]. Content providers spend considerable effort optimizing their infrastructure to deliver data from a server to the client as well as possible [2, 5, 10]. This high-volume traffic has been subject to several forms of traffic management for several years now. A commonlydeployed traffic management mechanism, policing, limits a flow to a preconfigured throughput rate, for example to enforce a bandwidth corresponding to a data plan purchased by a user, with any traffic exceeding the rate being dropped immediately. However, a traffic policer (the logical entity in a router or middlebox that performs policing) can often be configured to accommodate short bursts that exceed the rate 0.0 0.5 1.0 1.5 2.0 2.5