A Market for Scheduling, with Applications to Cloud Computing

ACO Student Seminar
Friday, November 20, 2015 - 13:05
1 hour (actually 50 minutes)
Skiles 005
Georgia Tech
We present a market for allocating and scheduling resources to agents who have specified budgets and need to complete specific tasks. Two important aspects required in this market are: (1) agents need specific amounts of each resource to complete their tasks, and (2) agents would like to complete their tasks as soon as possible. In incorporating these aspects, we arrive at a model that deviates substantially from market models studied so far in economics and theoretical computer science. Indeed, all known techniques developed to compute equilibria in markets in the last decade and half seem not to apply here.We give a polynomial time algorithm for computing an equilibrium using a new technique that is somewhat reminiscent of the ''ironing" procedure used in the characterization of optimal auctions by Myerson. This is inspite of the fact that the set of equilibrium prices could be non-convex; in fact it could have ''holes''. Our market model is motivated by the cloud computing marketplace. Even though this market is already huge and is projected to grow at a massive rate, it is currently run in an ad hoc manner.Joint work with: Nikhil Devanur, Jugal Garg, Ruta Mehta, Vijay V. Vazirani