Fast scheduling for Optical Flow Switching

• Main Author: Zhang, Lei, Ph. D Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, fl. 2014.
• Other Authors: Vincent W.S. Chan.
• Format: Others
• Language: English
• Published: Massachusetts Institute of Technology 2011
• Subjects: Electrical Engineering and Computer Science.
• Online Access: http://hdl.handle.net/1721.1/60828

Thesis (S.M.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 2010. === Cataloged from PDF version of thesis. === Includes bibliographical references (p. 97-99). === Optical Flow Switching (OFS) is a key enabler of future scalable optical networks. In the past decade, the OFS architecture has been studied to build an all-optical data plane to provide an end-to-end, cost-effective data transport to users with large transactions. Flow switching provides low-cost service to high-end users by relieving the IP routers on the edge of wide area networks from large transactions. However, the scheduling process of OFS presents possible queuing delays of several transaction durations. For some special applications with urgent time deadlines, the users want to bypass the queuing and pay more to use the network as soon as possible. In this thesis, we propose a fast scheduling algorithm which utilizes a probing approach to enable OFS to set up end-to-end connections for users with urgent transactions with a delay of slightly more than one round-trip time. A central control manager is used to periodically collect from network regions their most recent entropy evolutions of the network states and broadcast this information across the whole network in the control plane. With this information, fast setups of end-to-end all-optical connections for OFS are achieved by probing independent paths between source and destination, and reserving the available light paths along the way. A modified Bellman-Ford algorithm is designed to select the paths with the least blocking probabilities. By grouping details of network states into the average entropy, we can greatly reduce the information collected and disseminated by the centralized controller, making the network management and control system scalable to large networks. Since our algorithm makes no assumptions about network models or traffic statistics, it is robust against model variations, and any future changes in network topologies and traffic patterns. The technique can also be used in heterogeneous networks, in which networks from different domains are interconnected to provide a broader coverage. === by Lei Zhang. === S.M.