Grad Student Seminar
Server Allocation at Virtual Computing Labs via Queueing Models and Statistical Forecasting
The Virtual Computing Lab (VCL) is a cloud computing service that provides users remote access to software applications. The main challenge is to decide how many servers should be preloaded with which applications, and how many servers should be left flexible, to be loaded with the requested application on demand. If a preloaded server with a desired application is available, the user gets immediate access. If not, but a flexible server is available, the user gets delayed access after some extra loading time. If no active (on) server is available, he is blocked. We measure the service quality by the fractions of users who get immediate or delayed access, and the system cost by the number of on servers. To minimize the long-run system cost subject to a specified service quality, we propose policies that dynamically allocate servers in response to time-varying demand. We evaluate the policies with discrete event simulations using three-year data from the VCL of NC State University. We apply moving average and singular value decomposition based forecasting methods combined with two estimation methods to estimate model parameters. Our recommended policy achieves the target service quality using less than half of the servers needed under current policy.
Ship Motions in Irregular Seas
A broad look at some interesting problems related to ship stability and motions in irregular (random) seas, motivated by the real-world task of predicting rare but devastating events such as capsizing or parametric rolling. The talk will include descriptions of techniques from extreme value theory for inference regarding probabilities of rare events, as well as a discussion of the random oscillator model used to describe the physical dynamics of a ship in irregular seas and the benefit of approaching the system from both physical and statistical perspectives.