Algorithms for Dynamic Topology Optimization in Free Space Optical Networks The Maryland Optics Group Jaime Llorca, Aniket Desai, *Stuart Milner, and Christopher C. Davis Department of Electrical and Computer Engineering, *Department of Civil and Environmental Engineering Free Space Optical Networks • • • • • Free space optical networks are emerging as a cost-effective, rapidly deployable and portable technology, which provides secure, extremely high data rate communication. Civil Sector – Last mile problem – High quality video-based surveillance Military Sector – Dynamic Backbone: secure, high data rate communication Dynamic Nature – Atmospheric obscuration – Node mobility We present algorithms and heuristics for topology reconfiguration as a means to dynamically optimize network performance. The Autonomous Reconfiguration Process • Hybrid Battle-Space Networks • Airborne Free Space Optical Backbone Optical Links Terrestrial • • + Near-Surface, Surface and Sub-Surface user Systems Peer-to-Peer and Base-Station RF “subnets” • Minimizing Atmospheric Obscuration • • • The obscuration factor is a measure of the attenuation of the optical signal as it travels through the atmosphere, caused by the presence of clouds, fog, haze, rain, snow, etc. A cost measure is assigned to every possible link in the network as a function of the obscuration factor . The objective is to find a topology with minimum aggregate network cost. We require the network to be bi-connected, which assures the presence of at least one path between any pair of nodes, even after a single link break. Ring Networks • • • Every node has 3 optical transceivers. 3-MST + Reconfigurable Spanning Path (degree3 old in the graph): – Step 1: a MST with degree<=3 is built. – Step 2: nodes with degree<2 are connected using a simple path (reconfigurable spanning ring concept). 3-MST + Bridge Covering (degree3 new in the graph): • • • • Distribution of Topology Solution Deployment/Reconfiguration Multi-Hop Heuristic Single-Hop Heuristic • Single-Hop View: Choose an SD (s, d) pair and try to create a single-hop path by adding a new link between s and d. Higher traffic needs to be routed on fewer hops. Sort traffic matrix in decreasing order of traffic. Links are added in order of higher SD pair traffic until the topology is created. Time Complexity: O(N2) • • • 0 1 • Link-3 3 • Next R02, add a link between (0, 2) Next is R03. Interfaces at 0 are not available. Pass. 2 3 Next is R13, add the link between (1, 3) 1(s) 2 Link-4 Link-1 R01 = 100 MBPS, R02 = 90 MBPS R01 is maximum, add a link between (0, 1) Multi-Hop View: Interfaces not available between (s, d): Try to create multi-hop path. 0 R03 = 75 MBPS, R13 = 60 MBPS • • Solution Computation Key Idea: Arrange a network to evenly distribute the traffic, thus avoiding congestion or delays. Traditional Approach: Routing (TCP/IP). The topology is given. If a path is broken, nothing can be done. Assumptions: – Traffic matrix is known. – Every node has 2 transceivers. Ring topologies Objective: Minimize congestion. Congestion is defined as the traffic-rate on the maximally loaded link. – Step 1: a MST with degree<=3 is built. – Step 2: edges between nodes with degree<3 are added in order of greater bridge covering capability. – Achieves closer to optimal solutions in terms of cost minimization. Performance of Spanning Ring and Reconfigurable Spanning Ring heuristics Collection of Link State Information Minimizing Network Congestion 3-Degree Networks Every node has 2 optical transceivers. • Spanning Ring: MST based heuristic with degree constraint. • Reconfigurable Spanning Ring: additional processing provides more flexibility to find feasible topologies in adverse conditions. • Link State Examination Topology control involves a number of interacting processes that provide the network with autonomous reconfiguration capability. Local topology information needs to be dynamically gathered at a central node, Global network information is then processed, an optimized topology is computed, and the new network configuration is disseminated. In most cases, finding a topology which optimizes a given physical or network layer metric (obscuration, congestion, delay, etc) is an NP-complete problem. Heuristics have been developed to achieve near-optimal scalable solutions. • • Link-2 4(d) Link Selection: Local Optimization: Select Link that causes minimum increase in congestion in current cluster. Time Complexity O(N2) Performance: Better than Single-Hop heuristic. Drawback: R03 left unassigned by the algorithm. As topology finished, R03 may be routed on congested path. Scalability: both heuristics have polynomial time complexity, as shown in the following graph. Run-time of Spanning Ring and Reconfigurable Spanning Ring heuristics Performance of 3-degree heuristics • Scalability: Both heuristics have polynomial time complexity. Reconfiguration times range from 13us (8 nodes) to 50 ms (50 nodes). Run-time of Single-Hop and Multi-Hop Heuristics 5 Average Performance Improvement (Congestion reduction) of Multi-Hop over Single-Hop Heuristic