Online Mechanism and Virtual Currency Design for Distributed Systems

As distributed systems increase in popularity and experience resource contention, new resource allocation methods are needed for scalability and manageability. In this thesis, I investigate the use of auctions, a type of market-based methods, as a decentralized resource allocation approach. Agents in auctions individually submit bids that provide critical information, including agents‘ private value for resources, for prioritizing resource allocation. Three issues must be addressed for wider acceptance of using auctions. First, I present empirical data of a deployed market-based resource allocator, called Mirage, for allocating sensor network resources. The data provides observations of agent bidding patterns, which varied across values, sizes, and allocation timing. Furthermore, agents exhibited strategic behaviors that validate the need to mitigate such behaviors, which create complexity for both agents and system administrators. Second, I design Roller, an online mechanism that is strategyproof with respect to agents‘ submitted values and resource sizes. Roller is also configurable and able to provide different tradeoffs in regard to mis-reports of allocation timing. As agents of distributed systems often require responsive decisions, Roller uses a rolling window abstraction that enables the allocation of future resources. When compared to other allocators, Roller provides a high value and high responsiveness environment that is suitable for agents with iii