A study of the IEEE 1394 cluster and its application in parallel transportation simulation systems

The design of modern transportation systems leads to many challenges and it makes great impact on societies. Computer simulations of transportation system provide a mean to study the system performance or to determine the most effective solution for a problem. Parallel and distributed simulation methods are often used to reduce the execution-time of large-scale simulations and network of workstations or clusters become a popular method to speedup large-scale simulation systems.; Cost-effectiveness is the major factor that drives the use of computer clusters. In this thesis, a computer cluster is proposed, which employs a new interconnection device - the IEEE 1394, or Firewire, as the interconnection media. The IEEE 1394 is relatively low in price, while good in performance comparing to other interconnection medium. Such cluster is employed to implement two typical microscopic transportation simulation systems, the performance improvement of the two systems are observed, and therefore, these systems can accommodate large-scale simulations in shorter time.; The Firewire system is originally designed for interconnecting digital audio/video equipments to a computer and in order to utilize the device as a general purpose networking media, a custom-designed communication protocol is devised. The new protocol reduces the communication overhead and an improvement of around 25% is shown when comparing to the traditional TCP/IP protocol. In order to fully utilize the transmission capacity of the two operation modes, a fuzzy control mechanism is applied to the communication protocol. The fuzzy controlled communication protocol shows a speedup factor of up to 20% when comparing to the ordinary mechanism.; The performance of the IEEE 1394 cluster is compared to the clusters using other interconnection media such as the Ethernet to illustrate the differences in terms of cost-effectiveness. Case studies of two typical transportation simulation applications substantiated that these applications benefit significantly from such a cluster. Case studies in the thesis are carried out on a cluster, which consists of only four workstations. A statistical model is developed to estimate the performance of such cluster in a larger scale. The model shows that with a cluster that consists of 64 workstations, the performance is still acceptable.