CAREER: Algorithms for Services - Oriented Communications Networks

University Of Southern California, Los Angeles CA

Investigators

Abstract

The focus in modern computer networks has started to shift towardsservices and functionality rather than efficiency andoptimization. This calls for algorithms that combine user-centricperformance metrics such as fairness, availability, and latency withmore traditional objectives such as the total bandwidth cost or totalthroughput; it is desirable for these algorithms to provide formalguarantees in abstract models, as well as general rules of thumb forsystem developers. This research involves the study of severalproblems within the above framework.The first problem is to design efficient distributed algorithms forachieving approximate fairness in computer networks, and to studysimple mechanisms that induce convergence of selfish users toapproximately fair solutions. The second problem is to study routingissues in peer-to-peer systems; peer-to-peer systems are groups ofend-nodes that collaborate to implement new network services at theapplication level. Another problem is to study the packet-injectionrate at which common packet scheduling protocols such as FIFO becomeunstable.In an attempt to overcome the gap between theoreticians andpractitioners in this field, a new course titled "Algorithms forcomputer networks" will be developed, and the course material will bemade publicly available. Another course titled "Modern algorithmictechniques" will also be developed.