Abstract:
Another of the best things about research is when you have a student come up to you and tell you that something you were sure was true is, in fact, false. And it is even better when she can prove it!

My second talk will discuss a recent research in which we consider a single-server queueing system (GI/GI/1 queue) where the value customers obtain from service increases with their service time, but decreases with their waiting time. Examples of such a system include doctor's offices or amusement park rides: customers don't like to wait in line, but do want to spend a long time seeing the doctor, or on the ride. For such a system we show, surprisingly, that given a homogeneous customer population, the performance of the system actually improves by increasing the variability in the system by varying the service rate. This is true even if the service rule is static, i.e. even if the service rate must be assigned to customers independent of the state of the system. In other words, instead of treating all the customers the same, it is better if upon arrival customers are randomly assigned a service grade, without any knowledge of how many customers are waiting already, where different grades are served with different service rates. We will discuss the optimal way to determine the assignment probabilities and the service rates of the different grades, the intuition behind this result, and how much improvement is possible. We will again conclude with directions for future research.

This talk is composed of results from joint work with Ying Xu and Katia Sycara.