This project is developing ad hoc end-to-end congestion control techniques particularly adapted to mobile ad hoc networks (MANETs). In mobile ad hoc networks, frequent link failures and other random losses unrelated to congestion disqualify packet loss as a reliable congestion indicator and require ad hoc congestion control algorithms that must (1) accurately diagnose the real cause of a loss and (2) appropriately react to each type of loss. The research team is investigating an innovative direction to develop an accurate loss discriminator and an implicit continuous congestion measure that is based on packet loss and yields better performance. The key to solving these problems is to "de-randomize" congestion losses. This is achieved by marking packets with different discard priorities and enabling biased queue management at all intermediate nodes so that they may target and drop packets with the highest discard priority when congestion occurs. A judicious marking allows senders to implicitly estimate congestion. This project attacks theoretical and systems challenges and will enable the practical deployment of MANETs. In addition, such challenges will help attract and educate talented graduate students and retain the brightest undergraduate students.
