My research in the growing area of intersection between economics and computer science focuses on economic modeling using modern statistical learning techniques as well as in algorithmic mechanism design. I have worked to apply such ideas to understanding individual behavior in large systems such as cloud computing and computer networks.
Spot Price Dynamics in Cloud Computing
We introduce a model for spot prices in two-market (spot and on-demand) cloud-computing environments. Using this we are able to gain insight into cloud provider behavior and learn parameters of a nonlinear dynamical systems that allow for spot price prediction. These predictions can then be used to inform bidding between instances to reduce the monetary cost of parallelizable jobs.
Joint with Liang Zheng, Andrew Lan, Carlee Joe-Wong, and Mung Chiang.
Equilibrium-Seeking Congestion Control
Some recent proposals for congestion control have sought to justify their protocols as converging to the Nash equilibrium of a multi-player partial-information utility game in finite time. We propose a simple, distributed, gradient ascent-based protocol that provably reaches equilibrium in polynomial time and performs well on simulations.
Joint with Nikunj Saunshi; advised by Jen Rexford.
Please also see parallel but more-complete work of Dong, Meng, et al.
Incentive Schemes for Internet Exchange Points
We model incentive schemes for setting up local routing agreements and Internet exchange points in developing countries and examine the use of semi-definite relaxations and simulated annealing approaches to solve the resulting non-convex optimization problems. Since such setups incur high starting costs but greatly improve service, understanding these markets is important to understanding how best to expand Internet coverage.
Joint with Michael Chang; advised by Sanjeev Arora and Nick Feamster.