Kihong Park

  Associate Professor    |    Department of Computer Science    |    Purdue University

Research Contributions

Dr. Park's research contributions span several areas, with focus on Internet traffic modeling and control, noncooperative game theoretic approaches to network resource allocation, and protection against DDoS and worm attacks in power-law networks. The article, "The Internet as a complex system" (Chapt. 1, The Internet as a Large-Scale Complex System, Oxford University Press, 2005), gives an overview of the three areas.

Internet Traffic Modeling and Control

Following Leland et al.'s seminal discovery of self-similar Internet traffic [SIGCOMM '93], Park's contribution focused on explicating the structural causes of traffic self-similarity and its control. Park et al. [ICNP '96] showed that the principal cause of self-similar Internet traffic is heavy-tailedness of client/server file systems mediated by TCP. Although long-range dependence in self-similar traffic can adversely impact network performance, Park noticed that predictability at larger time scales can be harnessed for traffic control, for example, improving TCP throughput by 50-60% [TOMACS '00]. Park pioneered the area of multiple time scale traffic control, also serving as external Ph.D. committee member on two doctoral theses (S. Ostring, Univ. of Canterbury; G. He, Univ. of Illinois) in this area. This work continues to date as part of a DoD project that applies self-similar traffic predictability to ad hoc wireless routing (joint work with Jason Li at IAI). Representative publications:

Noncooperative Game Theory Approach to Network Resource Allocation

Park has studied the structural properties of multi-class network systems when viewed as a noncooperative game driven by selfish users. These games arise naturally in multiservice networks (e.g., IETF DiffServ) and TCP congestion control (a single-class system). The main challenge in understanding these systems lies in nonlinear coupling among users and service classes, a form of externality, induced by resource boundedness. Park et al. [ICE '98] give a comprehensive characterization of noncooperative multi-class service provisioning systems with respect Nash equilibria, Pareto and system optima. Chen and Park [INFOCOM '99] consider networks of multi-class switches, Ren and Park [IWQoS '00] study the impact of packet scheduling on stability and efficiency, and Lomonosov et al. [JCSS '04] incorporate the effect of pricing and computational complexity. Park has graduated two Ph.D.s (S. Chen and H. Ren) whose theses are on this topic. Representative publications:

Protection against DDoS and Worm Attacks in Power-Law Networks

Around the year 2000, a number of networking researchers entered into the network security arena, advancing networking solutions such as probabilistic packet marking and route-based packet filtering to Internet security problems that were previously dominated by cryptographic and infosec methods. Park and Lee [SIGCOMM '01] proposed route-based packet filtering as a proactive method for protecting against spoofed DDoS attacks. Park was one of the first to exploit the power-law connectivity of the Internet for network security, which parallels his earlier work that harnessed heavy-tailedness for self-similar traffic control. Park et al. [AFRL '04] extended distributed packet filtering to local containment of global worm attacks as part of a DARPA ATO FTN project. Park and Lee [INFOCOM '01] contributed to probabilistic packet filtering by identifying and analyzing several defense vulnerabilities. Park has trained two postdocs (H. Lee, Assist. Prof., Korea Univ.; A. Selcuk, Assist. Prof., Bilkent Univ.) in this area, and two Ph.D. candidates (B. Bethala and H. Kim) are completing their theses in network security. Representative publications:

Other Research Contributions

Park's research contributions in other areas include genetic algorithms where his work has shown the limitation of GAs when treated as a combinatorial optimization technique, cluster computing where he shows that congestion control can significantly speed up parallel/distributed computing applications by regulating network access, cellular automata (his Ph.D. thesis) where he shows that certain 1-D cellular automata proposed by Gacs, Kurdyumov and Levin (1978) and Toom (1980s) as candidates of non-forgetful systems are ergodic when subject to biased noise, and wireless networks where he shows that IEEE 802.11 WLANs are much more resilient to multiple access contention than previously believed. Representative publications:

Publications: Recent | Old