PC-Cluster simulator for joint infrastructure interdependencies studies

Rapid advances in network interface cards have facilitated the interconnection of inexpensive desktop computers to form powerful computational clusters, wherein independent simulations can be run in parallel. In this thesis, a hardware and software infrastructure is developed for the simulation of a complex system of interdependent infrastructures. A PC-Cluster is constructed by interconnecting 16 off-the-shelf computers via high speed Scalable Coherent Interface (SCI) network adapter cards. To enable synchronized data transfer between the cluster nodes with very low latencies, a special library comprised of communication routines was developed based on low-level functions of the SCI protocols. Different interrupt mechanisms for synchronous data transfer between the cluster computers (nodes) were investigated. A new method of implementing the interrupts is developed to achieve a 3.6 μs latency for one directional data transfer, which is shown to be an improvement over the standard interrupt mechanisms. To facilitate distributed and concurrent simulation of Simulink models on different computers of the PC-Cluster, a special communication block with the appropriate GUI interface has been developed based on the Simulink S-Function and interfaced with the developed SCI library. A reduced-scale benchmark system comprised of some of the University of British Columbia's infrastructures including the Hospital, Substation, Power House, Water Station, and Steam Station has been chosen to investigate the operation of PC-Cluster and potential improvement in simulation speed compared to a single-computer simulation. The models of considered infrastructures are distributed to different computer nodes of the PC-Cluster and simulated in parallel to study their interdependencies in a case of emergency situations caused by an earthquake or a similar disturbing event. It is shown that an improvement of computational speed over a single computer can be achieved through parallelism using the SCI-based PC-Cluster. For the considered benchmark system, an increase in simulation speed of up to 5 times was achieved.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat