My research interest lies in building protocols, architectures, mechanisms to support new uses of computer networks, and to measure, study, and characterize Internet topology and traffic. Most of the research here is conducted on a network testbed and simulation environment made possible by equipment donations from Sun Microsystems Inc. and Digital Equipment Corp. and housed in the UM Software Lab. Funds for this research is provided by the NSF CAREER Award, PECASE, NSF Special Project Awards, Office of Naval Research, and AT&T Labs-Research.

Prospective and current Graduate Students may also be interested in pointers to some useful resources I've collected.

I am currently involved with the following projects (in alphabetical order):

Internet Distance Map Service

It is increasingly the case that a given Internet interaction can be satisfied by one of a number of internet hosts. In any such interaction a major factor in choosing which host to access is the distance between the interacting hosts. Distance here refers to some internet performance metrics such as latency or bandwidth. The benefit of factoring in distance in the choice of which replicated web server or web cache to access is apparent. Of broader impact would be the use of distance information in the self-configuration of long-term peering relationships between servers providing network services, such as netnews servers, domain name servers, multicast routers, or web caches.

While hosts can measure characteristics of paths using various tools such as ping, traceroute, pathchar, mtrace, etc., having each host conduct performance measurements prior to each internet interaction inevitably leads to high overhead both to the host and to the internet. Hence a useful service for the internet would be one whereby a host could quickly and efficiently learn the distance between two other hosts. To be widely useful, such a service should provide an answer with a delay and overhead less than those of the gains achieved by using the service.

The objective of this research is to explore scalable design alternatives for an architecture to provide Internet Distance Maps Service (IDMaps). This is joint work with Paul Francis, Vern Paxson, Danny Raz, Yuval Shavitt, and Lixia Zhang.

My interest in performance characterization of network traffic started out with an empirical characterization of wide-area TCP/IP traffic. You can read about this work and the ensuing tcplib traffic characteristics library from my papers.

Internet Topology Characteristics

In this project, we consider the Internet as a graph of Autonomous Systems and are interested in the characteristics of the Internet's inter-AS topology, its spatial features as well as its behavior over time. (An Autonomous System (AS) is a network under a single administrative authority. ASs connect to each other through border routers; the Internet can be considered as consisting of interconnected ASs. Within an AS, the network could be further decomposed into subnetworks connected with internal routers.) As a side product of this part of the research, we expect to obtain models that are consistent with Internet-like topologies. The focus of our agenda, however, is on the more fundamental problem of investigating whether our models are in fact capturing the essence of certain underlying network design mechanisms or engineering constraints that result in random topologies that perforce exhibit many of the empirically observed phenomena. Thus, the research challenge here is to move beyond the traditional ``black-box'' approach to topology generation and the conventional approach of validating models against observed phenomena and instead try to understand and capture the possible causes that are responsible for many of the properties of the real Internet topology. To this end, we focus on two particular approaches that have recently surfaced in physics and complex systems literatures concerning the emergence of scaling phenomena associated with Internet-like graph structures or other complex, engineered systems.

More information, including papers and software, can be found at our topology web site.

Market-Based Resource Allocation

Traditionally, admission control algorithms make local decisions based only on local information. Preliminary results reported in the literature show that local admission decisions can lead to global deadlock condition or equilibria in which flows with certain characteristics become under-represented. No study to address these global dynamics issues has been reported. This research proposes to apply a market-based approach to global resource allocation. Market-based resource allocation mechanism (MBRA) allocates resources to flows within the framework of a price system. Flows compete for resources by their willingness to pay a higher price. The primary advantage of an MBRA is that its inherently decentralized nature allows for a naturally distributed solution. Furthermore, most research on network support for quality of service already assume the existence of a pricing infrastructure. While there is a sizeable body of literature in the application of market-based control to computing resource allocation, no previous work on distributed resource allocation that involves multiple resources and takes into account transaction costs, i.e. the cost to negotiate and execute a transaction, has been reported.

Multi-player Gaming

Online multiplayer games are poised to become a multi-billion dollar industry. With the advent of Internet-enabled gaming consoles and broadband access, multiplayer game sales could double in the next year. As the multiplayer game industry grows, so will demand for network protocols designed specifically for gaming purposes. We look at multiplayer gaming related network protocols, with the main emphasis on latency hiding and scalability.

Read more about our current project from the project page.