Network X-ities (design for "robust" network operation):
non-fragility, manageability, diagnosability, optimizability, scalability,
evolvability, adaptability, re-configurability, security,...
(please refer to
a talk given by
Professor Jim Kurose)
In order to build a rigorous foundation for a future robust and efficient
network architecture, I am currently investigating the following areas:
Modeling, design, and analysis of unstructured overlay networks
Traffic analysis in BitTorrent-like peer-to-peer networks
Combined routing and congestion control
Interactions among multiple overlay networks (structured/unstructured)
and underlay networks
Routing security
I am also investigating the selfish behaviors in wireless and sensor
networks.
Thesis research (part of Network
X-ities Project)
Interactions among self-interested users of network
resources
In recent years, some new applications (such as the
Resilient Overlay Network) indicate that self-interested users attempt to
strategically manipulate the state of the system to gain benefits for
themselves. These selfish behaviors might dramatically decrease the system
efficiency and cause instability to the network. Therefore, it is important to
understand how these selfish users interact with each other, and what impact
these interactions might have on the Internet. My research focuses on these
interactions. In the context of congestion control, I study whether the
interaction among selfish TCP users can collapse the Internet. In the context of
routing, I investigate the interactions between underlay and overlay routing
controllers by modeling them as non-cooperative players in a game. I also study
their interactions by investigating the ability of an overlay to compensate for
a “careless” underlay. I hope my research will provide a foundation for design
of efficient networks and other computer systems with robustness against selfish
behavior.
Can an Overlay Compensate for a Careless Underlay?
Honggang Zhang, Jim Kurose, and Don Towsley.
25th IEEE Conference on Computer Communications (INFOCOM 2006), April
23-29, 2006. Barcelona, Spain.
Also available as UMass CMPSCI Technical Report 05-49.[pdf]
TCP Connection Game: A Study on the Selfish Behavior of TCP Users. Honggang Zhang, Don Towsley, and Weibo Gong.
13th IEEE International Conference on Network Protocols (ICNP 2005),
November 6-9, 2005. Boston, MA, USA.[pdf]. A longer version available as UMass CMPSCI Technical Report 05-48.[pdf]
On the Interaction Between Overlay Routing and Underlay Routing. Yong Liu, Honggang Zhang, Weibo Gong, and Don Towsley. 24th IEEE Conference on Computer Communications (INFOCOM 2005), March
13-17, 2005. Miami, Florida, USA.
[pdf]
Measurement Study and Inference with IPID. The
16-bit identification field (IPID) of IP packet was investigated through
measurement studies to infer the amount of internal traffic generated by a
server, the number of servers in a load-balanced server complex, and the
difference between one-way delays of two machines to a target computer.
Multi-path Overlay TCP and TCP Relays.
Multi-path overlay TCP was designed and implemented to provide robust and
efficient transmission service in face of unpredictable Internet dynamics. In
addition, various algorithms were presented to optimally integrate application
level relays and transport layer control to dramatically improve data
throughput.
Differentiated Service Architecture with Feedback
Control. Proposed a simple and innovative scheme integrating edge and core
feedback controllers to provide guaranteed service for different classes of
traffic. Based on classic control theory, the edge controller was designed as
an adaptive token bucket (Adaptive Rate Management) and the core controller was a preferentially marking
Active Queue Management scheme.
Network Testbed. Built from scratch a testbed
consisting of six Linux machines, one Cisco Gigabyte Switch, and one FreeBSD
machine. Implemented a new Active Queue Management (AQM) algorithm in a Linux kernel and used this
customized Linux kernel as a router. Conducted extensive experiments to
compare this new AQM with Random Early Detect on the testbed and on the
Internet.
Active Queue Management with Adaptive Control. Proposed a
self-tuning structure that tunes the parameters of an AQM scheme as a function
of on-line estimates of network parameters. This structure was lightweight and
straightforward to implement. My control-theoretic analysis and simulations
showed that this scheme was remarkably robust to variations in network
conditions.
Experimental Evaluation of Active Queue Management Schemes for Highly
Variable Workloads. Developed a simulator to evaluate the performance of
several AQM schemes. The workloads were highly variable long-range dependent
traffic. Presented design recommendations for AQM to deal with highly variable
workloads.