E. Okorafor and M. Lu (USA)
: sequencing, performance, analysis, simulation, real-time, multiprocessor
Network performance for a single-hop WDM network deteriorates due to packet collisions that occur in both source and destination nodes. These collisions can also occur in both the control and data channels as in the case where reservation techniques are adopted. In order to reduce these collisions and maximize bandwidth utilization, we present in this paper an on-line scheduling method that can result in collision free communication and achieve real time operation in high-speed multiprocessor systems. The method we propose can be easily adapted to other network types employing optical interconnects. We conduct a throughput analysis of our scheme using methods from discrete-time queuing systems and use computer simulation results to compare the different variations of our scheduling algorithm. Our method efficiently utilizes the enormous bandwidth potential Optical Interconnects and Wavelength Division Multiplexing (WDM) techniques provide in computer network communication. This method is unique because it allows the transmission schedule to be staggered in time and space with provision for on-line packet arrivals and mixed packets types (unicast, multicast and broadcast). It is topology independent. Access to each data channel is controlled by a priority based cluster scheme that optimizes bandwidth utilization while minimizing packet delay. The basic idea of our proposed scheduling method is to show that the data packet sequencing has a profound effect on network traffic performance and, surprisingly, it has not received much attention. Accordingly, our results show that efficient message ordering in the data scheduling stage of the Routing and Wavelength Allocation (RWA) process, can better effectively utilize the enormous bandwidth potential available in optics. These results consequently lead to a remarkable improved throughput performance and lower average packet delay.
Important Links:
Go Back
