Gilbert (Gang) Chen

I am a Postdoctoral Research Associate with the Center of Pervasive Computing and Networking at Rensselaer Polytechnic Institute.  I got my Ph.D. in Computer Science from RPI in May 2003. Professor Boleslaw K. Szymanski was my thesis advisor and now is my postdoc supervisor. My current research includes simulation of wireless sensor networks and design of energy-efficient routing protocol for wireless sensor networks

Research

• Routeless Routing:  an energy-saving routing protocol for wireless sensor network.  Different from any other wireless routing protocol, this new routing protocol is free of route maintenance.  Consequently, it allows any intermediate node to turn off freely to conserve energy while new routes can be established instantaneously.
• Sensor Network Simulation: leading a team of 5 people developing SENSE, a new wireless sensor network simulator that are targeted at extensibilty, reusability, and scalability. Designed the core architecture and implemented the kernel of the simulator and some modules such as IEEE 802.11, AODV, and DSR.
• Parallel Discrete Event Simulation Using Lookback : this research was my Ph.D. thesis which investigates the use of lookback, the ability to change past locally, in enabling efficient synchronization protocols for Parallel Discrete Event Simulation.  Click the link for more information.

• COST:  a discrete event simulation tool.  What distinguishes COST from numerous other discrete event simulation software is that COST adopts a component-oriented world view that makes COST models highly reusable.  Written in C++ with a heavy use of templates, a COST model built for one simulation can be easily embedded in other simulations.
• SENSE:  a wireless sensor network simulator built on top of COST.  It features a very fast sequential simulation engine and an efficient memory management library designed exclusively for wireless networks.
• CompC++: a language extension to standard C++ that introduces the fifth programming paradigm, component-oriented programming (the existing four are function, modular, object-oriented, and template).  Most of SENSE and COST has been rewritten with CompC++, resulting in much clearer, shortened code (down from 1290 lines to 360 lines for cost.h), less compilation time (from 17.4 seconds to 8.3 seconds for SENSE), and greatly improved simulation speed (from 440,000 events/second to 720,000 events/second for SENSE).
• DSIM: a distributed optimistic discrete event simulation system that is capable of achieving an event processing rate of more than 5 million events per second and a speedup of 22 on 33 processors.

Teaching

• Spring 2003 CSCI-4320 Parallel Programming
• Spring 2002  Beginning C Programming for Engineers

1. Local Leader Election, Signal Strength Aware Flooding, and Routeless Routing , G. Chen, J. W. Branch, and B. K. Szymanski, accepted to the 5th IEEE International Workshop on Algorithms for Wireless, Mobile, Ad Hoc and Sensor Networks, Denver, Colorado, Denver, Colorado, April 2005
   
   Local Leader Election is referred to the problem of electing a leader among the neighborhood of a node.  A simple yet effective solution is presented.  The solution naturally leads to an entirely new routing protocol for wireless network named Routeless Routing, which possesses several interesting characteristics that were not intentionally targeted at.

    

2. Four Types of Lookback , G. Chen and B. K. Szymanski, in Proceedings of the 17th Workshop on Parallel and Distributed Simulation, p3-10, San Diego, June 2003
   
   After the publication of [4], we were still not certain if the notion of lookback had been understood correctly and thoroughly. It turns out that lookback defined in [4] is only a special form of general lookback. In this paper, we give the definitions of all four types of lookback, and show that all of them can be exploited in PCS (Personal Communication Services) simulation.
    

3. COST: Component-Oriented Simulation Toolkit, G. Chen and B. K. Szymanski, in Proceedings of the 2002 Winter Simulation Conference, p776-782, San Diego, December 2002
   
   This paper gives a detailed description to COST, our general purpose discrete event simulator. This simulation exemplifies the application of the component-based approach to sequential simulation. Modern C++ template-based techniques are adopted to implement the component-port model first presented in [7].
    

4. Lookback: A New Way of Exploiting Parallelism in Discrete Event Simulation, G. Chen and B. K. Szymanski, in Proceedings of the 16th Workshop on Parallel and Distributed Simulation, p. 153-162, May 2002
   
   Many simulation models are found to be capable of changing the simulation past locally, and this ability is named lookback. Traditionally, a dual of lookback known as lookahead, which is the ability to predict the future, was used in conservative PDES protocols, and it was widely believed that the performance of these protocols is bounded by the critical times of events. We proved in this paper that lookback is always more abundant than lookahead, and designed two lookback-based synchronization protocols that allow conservative simulations to circumvent the super-criticality limit, which was previously thought impossible by many researchers. Lookback can also be exploited in optimistic simulations to improve efficiency, by reducing the number of rollbacks and anti-messages.
    

5. A Component Model for Discrete Event Simulation, B. K. Szymanski and G. Chen, invited presentation, Lecture Notes on Computer Science, vol. 2328, p. 580-594, Springer-Verlang, Berlin, June 2002
   
   In [7], components are artificially divided into two worlds: conservative and optimistic. Adapters have to be used for components in different worlds to communicate with each other. We realized that this is not a good solution, and that the functionality classification should be imposed on ports, rather than component. In addition, the discovery of lookback in [4] helped us identify another type of ports, referred to as lookback ports.
    

6. Component-Oriented Simulation Architecture: Toward Interoperability and Interchangeability, G. Chen and B. K. Szymanski, in Proceedings of the 2001 Winter Simulation Conference, p. 495-501, Arlington, Virginia, December 2001
   
   This paper studies the implications of the component-based approach on the reusability of simulation models. We observed that the research community, especially the HLA people, is
   only concerned with the issue of interoperability. However, interoperability is just one form of reusability, with the other form being interchangeability. We presented a prototype of our CORSA simulation architecture that can guarantee both forms of reusability.
    

7. Component-Based Simulation, G. Chen and B. K. Szymanski, in Proceedings of the 2001 European Simulation Multi-Conference, p. 68-75, Prague, Czech Republic, June 2001
   
   This paper differentiates three types of simulation by the semantics of simulated time. Type I components do not have the notion of simulated time; Type II components notice the existence of simulated time, but cannot change it; Type III components maintain their own simulation clock themselves. Under this classification, all non-simulation programs or functions without a time variable are viewed as Type I components, while simulation models are of Type II. Type III components include sequential simulations where the simulated time is totally ordered, and parallel simulations where the simulated time is partially ordered. This classification captures the nature of software development involved in simulation systems, and enables a new component-based approach to hierarchically constructing large-scale simulations.
    

8. Linking Spatially Explicit Parallel Continuous and Discrete Models, B. K. Szymanski and G. Chen, in Proceedings of the 2000 Winter Simulation Conference, p.1705-1712, Orlando, Florida, December 2000
   
   This paper applies the mobile agent approach to the linking of continuous simulations and discrete event simulations. We argue that, the communication delay between distributed nodes is so large, that it is often beneficial to move the remote programs over the network to allow all participating programs to run on the same computer.
    

9. Multiparadigm Simulations in Modeling Spread of Lyme Disease, G. Chen, B. K. Szymanski, and T. Caraco, in Proceedings of the 2000 European Simulation Multi-Conference, Gent, Belgium, May 2000
   
   This paper introduces an approach to linking discrete event models with continuous models in the context of Lyme disease simulation. An interesting discovery is that, for linear systems, the property of superposition can be exploited to recover the current system state without calculating the exact past trajectory. This is, in fact, one of the primitive forms of lookback.

Postdoctoral Research Associate Department of Computer Science Rensselaer Polytechnic Institute 110 8th Street Troy, NY 12180  (last updated Nov 11, 2003)

Email: Homepage: http://www.cs.rpi.edu/~cheng3 Phone: (518)-276-2094(office)