Enabling Asynchronous Interaction and Coupling for Parallel Scientific Applications Using Decentralized Semantically-specialized Shared Spaces

OF THE DISSERTATION Enabling Asynchronous Interaction and Coupling for Parallel Scientific Applications using Decentralized Semantically-Specialized Shared Spaces by Li Zhang Dissertation Director: Professor Manish Parashar While large-scale parallel/distributed simulations are rapidly becoming critical research modalities in academia and industry, their efficient and scalable implementations continue to present many challenges. Key challenges are the dynamic and complex communication, coordination and coupling requirements of these applications, which depend on state of the phenomenon being modeled, are determined by the specific numerical formulation, the domain decomposition and/or sub-domain refinement algorithms used, etc., and are known only at runtime. Most existing solutions addressing these requirements are domain-specific and are typically tightly integrated with individual applications. Further, these solutions tend to be static and are not very flexible due to their underlying approaches. This dissertation builds on the Tuple Space model and customizes and extends it to support parallel scientific applications. Specifically, this research develops a semantically-specialized shared space abstraction infrastructure to support asynchronous interaction and coupling of parallel scientific applications. The key contribution of this thesis is a conceptual model and implementation architecture for realizing interaction and coupling frameworks that provide application-oriented abstraction for flexible, efficient, scalable, and asynchronous interaction and coupling. The approach emphasizes flexibility, efficiency, and scalability through the use of a tuple space-based abstraction which is customized based on application domain semantics. The approach and resulting systems complement and can be used in conjunction with existing