Synthetic Traffic Generation for Wavelet based Rosenblatt Process using LSTM for Multicore Architecture

Abstract

Synthetic traffic generation has become an important part for simulation on multicore architecture as it is more flexible than original raw traffic and also saves a lot of simulation time. A better simulation analysis helps the multicore architecture designer to choose more optimal use of hardware resources and helps in optimizing parameters such as energy, latency and throughput etc. A novel synthetic traffic generation algorithm is proposed for the simplest class of Hermite process which is non-gaussian traffic. The proposed algorithm uses the Long Short-term memory (LSTM) network for the accurate forecasting of inter-arrival times of the packets. It uses the two-dimensional architecture of time and space correlation of traffic system which consists of many memory cells. Analysis and results show the traffic generated from the forecasting technique is not different from the traffic generated from the statistical properties. It will ease the work of network designer to the optimal values of basic Hurst parameter for traffic generation by comparing the resource utilization parameters.