Study Of Key Technologies In Optical Grids

The rapid increase of network bandwidth and the emergence of new data intensive applications are the two driving forces for networking research and development. On the one hand, optical network bandwidth today has been expanded to 10Gb/s with 100Gb/s emerging, which enables many data intensive applications that were impossible in the past. On the other hand, new distributed applications, such as high-energy physics, earth science, astronomy, grid computing, expedite the deployment of high-speed wide-area networks. Under this background, investigations of architectures, protocols and algorithms of optical networks for these new applications, especially grid computing are attracting attentions both from academia and industry all over the world. This issue, referred to as optical grids, has a lot of key issues need to be addressed.Therefore, according to the types and features of grid applications, this dissertation investigates architectures of optical grids, resource discovery and management schemes and key technolygies in two network paradigms, i.e., optical burst switching (OBS) networks and automatically switched optical networks (ASON). Experimental and simulation results verify that the proposed novel architectures, protocols, algorithms and mechanisms can improve the performance of optical grids to a great extent.The main works of this dissertation are summarized as follows:1. Firstly, investigations of novel architectures for optical grids. According to the requirements, support of grid applications requires not only large bandwidth provided by optical networks, but also efficient architectures in which the resource discovery and management scheme is implemented. To address this issue, several novel architectures for optical grids are proposed in this dissertation, including a session initiation protocol (SIP)-OBS architecture, a peer-to-peer (P2P)-OBS architecture, a SIP/P2P/OBS hybrid architecture, a P2P/ client-server (C-S) hybrid architecture and an OBS/ generalized multi-protocol label switching (GMPLS) self-organized architecture. Based on a labeled OBS (LOBS) testbed and a Grid video-on-demand system which is implemented by Globus Tookit, these novel architectures are demonstrated and evaluated. Experimental results show that these architectures can avoid the disadvantages of the conventional C-S based architectures and can efficiently support future grid computing applications.2. Then, studies of serveral key technologies in OBS networks for grid applications. It is well known that the the classic TCP Reno protocol is inadequate to address the requirements of emerging grid applications on OBS networks. In order to find a most efficient TCP protocol in the scenario of Grid over OBS, three prominent protocols including HSTCP, Westwood and FAST TCP, are selected from three different categories of high-speed TCPs respectively. A modified theoretical model for TCP Reno over OBS is investigated, which is more accurate to model TCP Reno throughput compared with previous works and can evaluate the throughput of the aforementioned high-speed TCP protocols. By using the LOBS testbed with high bandwidth delay product, the modified theoretical model and evaluation results are verified through experiments. Moreover, since TCP Reno is the prevailing mechanism for data transmission in internet today, several novel OBS assembly and scheduling mechanisms are proposed to improve the performance of TCP Reno in OBS networks. Different from previous works, this dissertation focuses on TCP acknowledgement (ACK) packets. Considering the disadvantages of the traditional assembly mechanism and the characteristic of ACK, an independent ACK assembly (IAA) scheme and its extended mechanisms are presented. Simulation results show that these new mechanisms can improve TCP Reno throughput over OBS network significantly, which are satisfying the high available bandwidth requirements of future data intensive applications.3. Finally, investigations of serveral key technologies in ASON networks for grid applications, especially grid job congestion and how to improve the network efficiency. The grid job blocking in ASON is analyzed in detail and a modified algorithm, which is based on erlang reduced load approximation, is proposed to calculate grid job blocking probability in ASON. Differnet from previous works, the distributed call and connection management recommended by ITU-T and the signaling delay time are taken into consideration in the analysis and the algorithm. Simulation results show that the proposed algorithm is more accurate than previous works. After that, this dissertation verifies that ASON and the conventional admission control mechanism have serious problem to support future large-scale Grid computing. In order to address this issue, a novel dynamic call and connection admission control (DCCAC) scheme is proposed to improve the network performance and guarantee the quality of service of Grid applications. This scheme is applicable with complete network information and no network information. Numerical results show that DCCAC can improve the efficiency of the network to a great extent.