| Distributed systems are widely used in people's daily life. There are different resources in the distributed systems, including computing processors, memories, bandwidth, storages, etc. When tasks are processed in distributed systems, workload should be well scheduled among computing resources. Dynamic load balancing is the main goal of many runtime schedulers to adjust workloads on the fly.;This thesis proposes different resource management mechanisms for different distributed systems. First, a new scheme of workload classification is introduced. Then, the virtual memory management in clusters is discussed. A static computation assignment tool, Pitcher, is developed later. After that, a novel algorithm for load balancing on mesh networked computers is presented. At last, parallel computation distribution strategies in heterogeneous architecture machines are proposed and compared. |
