Many mathematical models have been proposed to evaluate the execution performance of an application with and without checkpointing in the presence of failures. They assume that the total program execution time without failure is known in advance, under which condition the optimal checkpointing interval can be determined. In mobile environments, application components are distributed and tasks are computed by sending and receiving computational and control messages. The total execution time includes communication time and depends on multiple factors, such as heterogeneous processing speeds, link bandwidth, etc., making it unpredictable during different executions. However, the number of total computational messages received is usually unchanged withinan application. Another special factor that should be considered for checkpointing purpose is handoff, which often happens in mobile networks. With these observations, we analyze application execution performance and average effectiveness,...
Xinyu Chen, Michael R. Lyu