A dynamic and decentralized approach to management of CPU and memory
Avraham Leff · 1992
This thesis proposes a dynamic and decentralized approach for resource management in distributed systems. This approach can be applied independently at each site in the system; as a result, resource management does not require a centralized coordinator. Sites cooperate in transmitting important state information to one another. Decisions made at one site are then factored by other sites into subsequent decisions. The approach is applied to two resource management issues. The first, load sharing strategies, attempts to use the system's overall CPU power efficiently. The second, object replication strategies, attempts to use the system's overall memory resources efficiently. Various algorithms are proposed and compared against the performance of alternative algorithms. Simulation results validate the effectiveness of this approach.