Study Of High Speed Traffic Transmission In Optical Burst Switching Networks

Over the past few years it has become evident that local computational resources cannot keep up with the demands generated by some grid applications such as particle physics, high-performance computing, visualization and e-health. Network researchers have been looking for a suitable infrastructure. Optical Burst Switching with the advantages of flexibility, low latency, and high bandwidth utilization, was considered to be meeting the above-mentioned high performance application. However, the investigation on how to efficiently carry upper layer high performance applications over OBS networks has just come into sight; a lot of key issues are still under studied.Therefore, for satisfying the requests of grid applications, the architecture of OBS grid which was capable of providing absolute QoS guarantee was proposed out. Consequently, we presented a simple method to configure high bandwidth large delay OBS test platform. Based on this test platform, the performance experiment of some TCP variants were reasonably evaluated and validated.Following are details:1. The dissertation firstly interprets why the grid application can be better supported by OBS Grid network. On this basis, the architecture of OBS grid which was capable of providing absolute QoS guarantee was proposed out and its processes of QoS management were discussed in detail. Simultaneously, delay jitter performance of OBS network was experimentally tested. The results proved that, without considering the impact of deflection routing, in order to absolutely distinguish delay jitter, the influence of OBS assembly time threshold and network traffic must be seriously considered. At the same time, theoretical analysis shows that the use of class-based weighted fair queuing and the priority preemptive LAUC-VF algorithm was enable to absolutely distinguish end-to-end delay, bandwidth and burst drop ratio of OBS network.2. Then, the transmission performances of high-speed and long-distance data were experimentally studied over OBS network. The experimental results of loss-based TCP variants showed that, in high-speed long-distance OBS network, under different burst drop ratios, HSTCP got the best throughput performance, followed by STCP and BICTCP. From the test results, it is obvious that the throughput fell most rapidly when the burst drop ratio increased from 0.2 percent to 0.4 percent. Once the burst drop ratio was greater than 0.4 percent, their throughput would decrease and below the half of available bandwidth. The further investigation confirmed that STCP show the highest sensitivity, followed by BICTCP and HSTCP.At the same time, the experiment of delay-based TCP variants showed the Fast TCP was not suitable for using on the OBS network. Since the burst data was transmitted by all-apical path in the OBS core nodes, there is no queuing delay on the inner network of OBS. Despite the subsequent experiment displayed the TCP Vegas mechanism has similar throughput performance compared to the HSTCP, STCP, as well as BICTCP, but it obtained the poorest sensitivity. Subsequently, ACK assembly effect was firstly found in the throughput experiment of HSTCP. The experimental results validated the throughput of HSTCP was primarily impact by the ACK-side assembly time threshold of the OBS edge node.3. At last, technique of pacing was deployed in the HSTCP and its pacing performance was tested in long-distance OBS network. Experimental results proved the more obvious performance improvement was achieved by using HSTCP pacing when the probability of network congestion increased in OBS network. Properly utilizing the technique of pacing will greatly enhance bandwidth utilization of HSTCP while burst drop ratio of OBS network was high enough. Moreover, it also reduces the average RTT of HSTCP and improves the stability of HSTCP. The experimental results also demonstrated that, under the 1.6 percent burst drop ratio, appropriately selected pacing period, the stability performance of HSTCP would be enhanced more than 70 times when compared to the performance of non-pacing and the throughput of HSTCP wasn't significantly change as well as.