Publication Details
Polykarpos Thomadakis, Christos Tsolakis, Konstantinos Vogiatzis, Andriy Kot and Nikos Chrisochoides.
Published in AIAA Aviation Forum 2018, Atlanta, Georgia, June, 2018
Abstract
This paper presents the design and implementation of the new version of our software, the Parallel Runtime Environment for Multi-node Applications (PREMA 2.0). The motivation for the development of the runtime system is to provide a software framework that efficiently handles irregular parallel applications whose performance suffers from work-load imbalances. The burden of monitoring the data and computation on the application level can be heavy and difficult to manage on a complex application like parallel mesh generation and error-based adaptive Computational Fluid Dynamics (CFD) solvers. The runtime system provides the minimum set of tools for developing applications that do not explicitly handle load balancing, by monitoring the system load and making dynamic load balancing decisions implicitly, based on user directives. Namely, PREMA 2.0 supports multiple execution units allowing the application to run multiple computations in parallel, while it monitors the load of the system and performs migration when desired without interrupting execution. Concurrent migrations from the same process are also made available, giving flexibility to the load balancing algorithm designers. Furthermore, the communication and the execution have been separated into different threads to provide asynchronous message reception and instant computation execution at the arrival of new work requests. Progress is guaranteed on the system level without any involvement of the application code; once a request for execution is issued, it will be handled at some point before termination. Testing the system on an application of parallel mesh generation indicates an overall performance improvement of about 20 percent, by retaining a better work load distribution among the execution units.
