REACTIVE METHOD FOR EXECUTING PARALLEL APPLICATIONS IN DISTRIBUTED COMPUTING PLATFORMS.

摘要

The present disclosure is related to a reactive method to run parallel applications in the form of a DAG on distributed computing platforms. The distributed computing platform is distinguished by having geographically distributed processors, heterogeneous and not dedicated to the application. The main feature of the method to deal with the changing nature of computing resources (even when there is a failure), is through the cyclic execution of two main components: a method of cyclic assignment of tasks to processors and a method to tolerate failures in a processor. To achieve the above, the method considers detailed information of the parallel input application, of the progress over time of the tasks that make up the parallel application and the monitoring of the availability of computational resources (processors and communication links) that make up the distributed computing platform. The method guarantees, in most cases, the parallel application to end the execution.