Efficient Checkpointing for Heterogeneous Collaborative Environments: Representation, Coordination, and Automation.

Abstract

Checkpointing can be used to adapt resource utilization in heterogeneous distributed environments. In checkpointing, the state of a process is captured and later restored on a computer to restart execution from the point where the state capturing had occurred. Such capability can be applied to process migration for which resource utilization is adapted toward high-performance by moving a running process from one computer to another. For a heterogeneous environment, problems in checkpointing can be categorized into three domains regarding mechanisms to capture and restore the execution state, memory state, and communication state of a process. Although a few solutions have been proposed, a well-defined solution is not yet exist. This thesis presents a practical solution to capture and restore the process state in heterogeneous distributed environments. The solution is based on three novel mechanisms: the data transfer mechanism, the memory space representation model and its associated data collection and restoration mechanisms, and the reliable communication and process migration protocols. These mechanisms define the machine-independent representations of the execution state, the memory state, and the communication state. They work in coordination to perform process migration in a heterogeneous environment. A software system is designed and implemented to automatically migrate a process. A number of process migration experiments are tested on sequential and collaborative processes. Experimental results advocate correctness and practicability of our solution.