A Conceptual Model for Grid-Adaptivity of HPC Applications and Its Logical Implementation Using Components Technology

摘要

Today grid middleware is complex to be used, the development of grid-aware applications is error-prone and the Virtual Organization grid paradigm contrasts with traditional High Performance Computing (HPC) optimisation strategies relying on resource stability and known cost models. The authors analyse several aspects of grid adaptivity, and identify 4 roles: the active resource/execution manager, the proactive resource administrator, the reactive quality-service coordinator, the passive resource coordinator. They present a hierarchical model for a component-based grid-software infrastructure in which the resource administrator and the resource coordinator roles are assigned to grid middleware and the quality-service role to HPC skeletons. Roles interactions through interfaces are described for a component based infrastructure implementation. The resource administrator mimics functionalities of components containers of service-oriented architectures. The resource coordinator manages the life cycle of sets of processes over a pool of grid resources. It offers to upper infrastructure layers a Virtual Private Grid facade, simulating a processor cluster facility.