Managing Shared Resources in the Data Center Era: Computer Architecture Meets Game Theory
Resource sharing is vital to improving efficiency and amortizing cost in high-performance computer systems. Within modern data centers, users selfishly pursue individual performance without regard for others or the system. To address this challenge and study the strategic behaviors of self-interested users, I turn to algorithmic economics and game theory. In my thesis, I rethink resource management in computer architecture and systems, constructing mechanisms that are robust to strategic behavior. First, I describe novel methods to manage shared power supplies in modern data centers. Second, I introduce a processor bidding mechanism that operationalizes Amdahl’s Law to guarantee proportional shares. Finally, I demonstrate dynamic mechanisms that allocate shared resources across time.