Spring 2001

This Semester's Chats: [ 4/9 | 4/2 | 3/26 | FRI 2/16 12:00 | THURS 2/15 12:30 | THURS 2/7 11:30 ]

4/9/01 - Network Design Using Simulation Tools
Jeffrey Stone
Department of Mathematics and Computer Science
Dickinson College

The purpose of this research was to gain an understanding of the processes involved in computer network design, and to learn how computer simulation can used to save time, effort, and money before flawed designs are implemented. Using a small, fictional business and guiding principles of current network methodology and tools, two distinct Ethernet network "scenarios" for the business were designed, including details such as the number of servers, hubs, switches, et al., and how they would be connected.

The Extend ä simulation tool was used to model my "scenario" networks. Hierarchical Extend ä blocks were constructed to simulate various network elements (hubs, switches, servers, workstations) and Extend ä models were constructed from these blocks. Specifications from (then-) current network devices were used to set the numerical parameters of the simulated devices. The models were run and numerical reports were analyzed to determine such things as node latency, switch congestion, and packet response times. It was from this comparative analysis that a preferred design was determined.

A "one-year-later" scenario was then performed on the winning scenario to show the affects growth may have on the network. By showing the network performance after the company had experienced growth, the goal was to show whether the preferred solution was a good solution for the long term. Finally, suggestions were offered to improve the chosen network's performance.

4/2/01 - How businesses adapt their computer systems to the challenges of e-business and the economy.
Jim Bierman
Dickinson Class of '70
CFO Quintiles Transnational Corporation

3/26/01 - Some Statistical Aspects of the Performance of Biometric Identification Devices
Michael Schuckers
Department of Statistics
West Virginia University

Biometric Identification devices (or simpley biometric devices) were once considered to be purely in the realm of science fiction. These devices which take a measurement of a physiological feature, e.g. fingerprint, iris, hand geometry, and compare that measurement to an earlier stored template are becoming increasingly prevalent. You can now buy a commercially available laptop with a fingerprint reader build-in. The United States Department of Defense will spend several million dollars in the next decade to test and implement biometric technology. Though much work has been done in this area, standards and protocols for most aspects of these devices are under development but are not yet uniformly applied throughout the industry. One of the most important issues facing this field is the quantification of the matching performance of a biometric device. Matching performance is the devices ability to determine that you are who you say you are. In this talk I will discuss matching performance and discuss some of the statistical issues involved in estimating the matching performance of a biometric device.

2/16/01 - GeomNet: Bringing Graph Drawing to Users
Stina Bridgeman
Department of Computer Science
Brown University

In the first part of the talk, I will introduce graph drawing, its applications, and some approaches used in graph drawing algorithms. I will then present GeomNet, a distributed computing system making graph drawing algorithms easily available to users over the Internet. The goal of GeomNet is to reduce the obstacles faced by users not in the graph drawing community who need to visualize graphs.

2/15/01 12:30 - Simulation of Mobile Servers on a Congested Network
Barry Lawson
Department of Computer Science
The College of William and Mary

We consider a simulation-based network model to determine the home location of mobile servers in a congested network. A direct application of this study is to determine the optimal placement of EMS sites within a city by minimizing the expected response time for emergency requests. With random request arrival times, response times, and service times, we investigate the placement of servers on a fixed network to minimize expected response and service times. In addition, we compare results produced using two distinct server response paradigms.

2/7/01 11:30 - How to Improve the Efficiency of Your Programs without Even Trying
Patricia Johann
Assistant Professor
Department of Mathematics and Computer Science
Bates College

In the modular style of programming, solutions to larger computational problems are constructed by solving, and then appropriately combining solutions to, smaller such problems. The modular style of programming is associated with greater reliability, clarity, mainatainability, and reuse of software components. Because it is also associated with increased programmer productivity, modular program construction is an integral part of any reasonable software engineering process.

Unfortunately, modular programs also tend to be more inefficient than their monolithic counterparts. As a result, automatable techniques for transforming modular but inefficient programs into more efficient, but equivalent, monolithic ones are needed.

Fusion is a technique for increasing the efficiency of functional programs in precisely this way. In this talk we introduce short cut fusion, a particular technique for automatically fusing functional programs. After indicating how short cut fusion works (by applying it to a small example program), we consider the problem of measuring the effectiveness of short cut fusion and related techniques in practice (i.e., on "real" programs). We discuss the progress made toward this goal in the undergraduate research experience we hosted last summer, and conclude the talk by indicating directions for further research along these lines.