Paradoxes in Performance Optimization of Distributed Systems

|Re ecting the way in which a distributed system is used, we have di erent levels of distribution in the decisions on performance optimization for the distributed system, ranging from the completely centralized to the completely distributed. We identify such paradoxes that adding capacity to the system may degrade the performance of all users in the optimum of some intermediately distributed decision scheme. We characterize the cases where such paradoxical performance degradation occurs. We show that in some situations the magnitude of such performance degradation can be unlimitedly large. We discuss the above paradoxes on the models of distributed computer systems, and show numerical and theoretical results.