Application Traffic Modeling and Optimization for NoC Communication

Abstract

The present study aims to develop a simplified methodology for the generation of Synthetic newlinetraffic for multicore architecture. To begin with, self-similarity processes are studied for thegeneration of synthetic traffic. The exponent known as the Hurst parameter is calculatedby various method from Rescaled-range method to Whittle s Estimator used to representthe self-similar process. The synthetic traffic is used to simulate on the Mesh architectureof diand#64256;erent scalable sizes to decrease the simulation time and it provides the and#64258;exibility tonetwork designers. The synthetic traffic is generated for the advanced class of MPEG-4videos with HEVC. The algorithm presented here requires less time for its generation asits complexity is less as compared with other existing algorithms for the generation ofSynthetic traffic.The synthetic traffic is simulated on diand#64256;erent classes of traffic patterns such as Complement, Neighborhood and Uniform patterns which constitute the selection source anddestination pair. Some of the important parameters such as end-to-end latency, linkutilization, packet loss probability are calculated based on the simulation outcomes helpsin understanding the communication paradigm and helps the network designer for a betterselection of communication resources in the early design process of Networks-on-chiparchitecture. newline newlineSubsequently, the role of non-Gaussian processes is studied with the Hermite process. In newlinethis study, the role and requirement of Non-Gaussian process are identified, various class newlineof processes identified in Hermite process, the Rosenblatt process is one of them whose newlinedistribution is non-Gaussian. The algorithm designed for generating Non-Gaussian process newlinebased on Rosenblatt process whose function is the integral of time and space domain newlineconcerning Fractional Brownian Motion.The generation of Gaussian and Non-Gaussian process are compared with RosenblattNon-Gaussian process shows the burstiness in dataand#64258;ow is low in Gaussian and Non-Gaussian process as compared with Rosenblatt Non-Gaussian