English

Abstract

In recent times, Optical burst switching (OBS) is a raising contending technology in the network performance area. Therefore, because of the in efficiency of buffers, losses exist due to contention on the concurrent burst arrival in the core nodes. In this situation, contention loss occur which doesn't indicates the congestion in network. As a result, classification or differentiation of losses is needed in various applications in order to avoid the false congestion identification. In this paper, improved Loss classification technique which includes combined supervised and unsupervised learning technique is proposed for Optical burst switching networks. The proposed classification technique includes dynamically weighted HMM (D-WHMM) with Expectation maximization clustering algorithm. Priorly, differentiation between congestion and contention losses is observed from various burst. The resultant differentiation is applied to proposed improved classification method. D-WHMM is a supervised learning approach which observes the losses and classifies them into clusters. By this classification the differentiation between contention and congestion loss is done. Experimental result of proposed system provides improved and efficient TCP performance over Optical burst switching (OBS) networks.