Study And Implementation Of The Prototype Of Supporting Environment For Parallel Simulation Based On HPC

With the continuous development of the scale and complexity of simulation applications, on one hand, more complex entity models require higher computation ability which gradually goes beyond any single processor nowadays, on the other hand, increasing cooperation between entities within different hierarchies, different advance progress of various processors and synchronization of simulation time results in restriction of advance from slower nodes to quicker nodes, which lower the runtime efficiency of entire simulations. To improve simulations' speed-up, besides enhancing the processing ability of computation platform, the potential parallelizability of simulation task itself is worthy of exploiting. Moreover, parallel processing technique should be adopted, in order to advance the simulation on all nodes as soon as possible and to reduce the penalty (i.e. processor stalling) from interactions between entities. Therefore supporting techniques for parallel simulation based on HPC has become a hotspot issue within the computer simulation research domain.Aiming at the requirement of large-scale simulation applications, especially of closed loop advancing species, such as campaign analysis, system countermine and study of warfare problem, evaluation of the efficiency of battle scenario and weapon equipment, based on the absorption of merits of distributed interactive simulation technique, this paper explores the key techniques of the supporting environment for parallel discrete event simulation (PDES) based on HPC, such as the architecture, the modeling framework, the optimism time management and the rollback framework. The key research work and innovation of this thesis are generalized as following:1. After an in-depth research of the evolution history of the architecture of supporting platforms for simulations and the standardizing simulation architecture (SSA), considering the characteristics of high performance parallel simulation, a hierarchical architecture of supporting environment for p arallel s imulation b ased o n H PC i s p ut forward. T he architecture follows a bottom-up software development fashion, which makes it possible that the functions of higher layer can be built upon the functions' abstraction of lower layer.2. In allusion to the multiprocessors in HPC, a tentative plan of the modeling framework of parallel simulation is given by studying the traditional approach of describing, decomposing, distributing and scheduling PDES tasks. Following the object-oriented design idea, the framework provides users with fundamental building blocks, such as simulation object (SimObj) and event, in order to support rapid construction of applications.3. Considering the embodiment of conservatism of the Breathing Time Buckets Algorithm is the deferred post of event-scheduling message and the post of message lies on the determination of local event horizon, an adoption of the data structure of linked priority queue is put forward to organize the collection of simulation objects within a node, the collection of pending events of a simulation object and the collection of unreleased message of a node. Consequently, the efficiency of searching events and messages with a node has been improved...