> Projects > IDS

IDS (Intelligent Demand Shaping)

Bandwidth Allocation

IDS allows ISPs (Internet Service Providers) to allocate bandwidth to their consumers in a fair and efficient manner. At times of peak traffic, ISPs are increasingly experiencing a capacity crunch, forcing them to reduce users' bandwidth. Thus, they need to allocate their available capacity. We propose a way for ISPs to allocate available bandwidth in a fair and efficient manner, taking into account bandwidth allocations and usage over time.

ISPs already use proportional fairness to schedule sessions as they arrive at a base station. This process, however, generally ignores which users initiate which sessions: heavy users that continually initiate sessions and hog the network can still receive large amounts of bandwidth, even though their traffic significantly detracts from other users' experience. We propose a way to account for users' overall network demand, giving higher priority to those users either use less, thus taking up less bandwidth from other users; or who pay more to receive a better priority. At times of congestion, the ISP allocates bandwidth to active sessions according to these priorities, as well as the individual session sizes. Our algorithms are inspired by recent EDGE Lab works on fairness and account for these factors in a Nash bargaining/proportional fairness framework.

User Client

To enforce this bandwidth control from the user side, we develop a client that can control the bandwidth of different applications. We use receiver-side TCP advertisement windows to control the bandwidth rates, and offer users the option of manually controlling this bandwidth should they wish to do so. Our IDS client, currently implemented on the Windows platform (laptop and tablet), also includes a user interface that shows the usage and bandwidth allocation history for each application. With this interface, users can manually adjust the bandwidth of different applications, educate themselves about their usage of different applications, and see their predicted usage for the rest of the month.