Abstract

The Internet gives us access to many agents that can help us complete tasks. These agents can come either in the form of machines (cloud computing) or people (crowdsourcing). However, generally these other agents do not provide their services for free; instead, they have privately known costs for providing their service. I will show how auction theory, and more generally the theory of mechanism design, in some cases allows us to allocate the tasks efficiently in a way that benefits all (even if agents are strategic), whereas in other cases there are fundamental limitations. Specifically, in the context of mechanism design for scheduling unrelated machines (as proposed by Nisan and Ronen), I prove a lower bound of 1+\phi on the approximation ratio obtainable by such mechanisms. I also prove several other results for characterizing truthful mechanisms. These involve the geometry of such mechanisms and techniques that allow us to transfer characterization results across domains. I further study the problem where a task (or multiple unrelated tasks) must be executed, and our objective is to minimize the expected sum of the agents’ processing times, However, each agent does not know exactly how long it will take him to finish the task; he only knows the distribution from which this time is drawn. I present the ChPE mechanism, which is uniquely tailored to our problem, and has many desirable properties including: not rewarding agents that fail to finish the task and having non-negative payments.