Self-Similarity Analysis Of Networks On Chip Designed For H.264 Video Decoding

In the many-core era, the traditional bus architecture has been unable to meet the communication requirements of cores. But Network on Chip solved the bottleneck problem of power, global clock synchronization, latency, scalability etc. appeared in the traditional bus architecture, and will gradually replace the traditional bus architecture to carry out the communication task.To reduce the risk of design and ensure products to meet the performance requirements, modeling and simulation of Network on Chip becomes particularly important. Modeling and simulation mainly assesses performance of Network on Chip, which depends on the proper traffic model. FBM burst traffic model is used in thesis to generate different levels of traffic, with the study of the performance of Network on Chip, the results showed that the higher the burst nature, the worse the performance of Network on Chip. At the same time, due to noticeable differences of Network on Chip traffic characteristics on different applications, and therefore a study of Network on Chip self-similarity has been done based on specific application, It is helpful to establish a more accurate traffic model, and is more conducive to improve the performance of network, and this paper has studied the self-similarity characteristics of Network on Chip by selecting H.264 decoding application.In the paper, using SystemC language built a network simulation platform of 2D-Mesh Network on Chip based on switching mechanism of wormhole routing, and had a parallel partition of H.264 decoding algorithm, and mounted it to Network on Chip, and realized the function of visual configuration, traffic collection and performance index extraction, and finally completed the design of simulation system, and verified its functionality.Then, studied the traffic characteristics of H.264 decoding, analyzed the characteristics of traffic that obtained from the simulation system by inputting a variety of compressed video stream, the results show that the traffic of Network on Chip based on H.264 decoding is self-similar.Finally studied the influencing factor of traffic self-similarity, the results show that mapping scheme and routing algorithm both have influence on self-similarity of traffic. A good mapping scheme can effectively reduce the value of traffic burst and the value of burst decreases with odd-even routing algorithm when the network is congested. It is helpful to reduce the value of burst and improve performance of Network on Chip, When design the architecture of Network on Chip according to the specific situation to choose the appropriate configuration.