Solutions to Parallel and Distributed Computing Problems: Lessons from Biological Sciences

6.7 SUMMARY

In a mixed-machine, distributed heterogeneous computing environment, there is a suite of high-performance machines with different computational capabilities. These machines are interconnected by high-speed links. Such a suite of machines can be used to execute a single application, whose subtasks have diverse execution requirements, or to execute a meta-task, which is a collection of independent tasks with different computational needs.

For the single-application case, subtasks are assigned to and executed on the machines that will result in a minimal execution time for the overall task, considering subtask computation time and intermachine communication overhead. A GA-based approach was described for matching subtasks to machines and scheduling the execution of the subtasks. This approach is used off-line and is based on expected computation and communication times for the subtasks. A dynamic-parameter-sampling-based method for using this approach on-line in certain application domains was also presented. This on-line adaptation may be helpful when computation and communication times can vary significantly from the expected values, depending on the input data being processed.

For the meta-task case, the goal is to match each task to a machine in the suite so that the execution time for the entire meta-task is minimized, In the situation considered here, the matching process occurs off-line and plans the machine assignments and execution schedule for a meta-task for ...