Configuring Low Cost Metanetworks on A Shared Substrate

In a diversified internet, meta-networks (“metanets?for short) share a common substrate and offer value-added services to millions of users around the globe. Therefore, configuring low-cost metanets with links having enough bandwidth to accommodate all anticipated user traffic is critical to the success of the metanets. In this paper, we propose a novel pruning algorithm that configures metanets on any given substrate in a cost-efficient way. In contrast to other testbed configuration systems, we solve the metanet configuration problem from a higher level specification and produces a network that is dimensioned to handle the specified traffic. To the best of our knowledge, our work is also the first one that tries to automatically determine the best metanet topology while considering network switching costs and propagation delays. We study how the best topology changes on different substrate networks as traffic conditions vary. In general, we find that as pair-wise traffic constraints and delay bounds are relaxed, the least-cost metanet topology becomes increasingly “tree-like? We also show the impact of delay bounds on the network costs under different traffic conditions. Our algorithm produces metanet configurations that are demonstrably close to the computed lower bound and is fast enough to handle substrate networks of practical size. Type of Report: Other Department of Computer Science & Engineering Washington University in St. Louis Campus Box 1045 St. Louis, MO 63130 ph: (314) 935-6160 Configuring Low Cost Metanetworks on A Shared Substrate Jing Lu and Jonathan Turner Department of Computer Science and Engineering Washington University in St. Louis St. Louis, MO 63130 Email: {jl1, jst}@arl.wustl.edu Abstract— In a diversified internet, metanetworks (“metanets” for short) share a common substrate and offer value-added services to potentially large numbers of users around the globe. Therefore, configuring low-cost metanets with links having enough bandwidth to accommodate all anticipated user traffic is crucial to the success of the metanets. In this paper, we propose a novel pruning algorithm that configures metanets on any given substrate in a cost-efficient way. In contrast to other testbed configuration systems, we solve the metanet configuration problem from a higher level specification and produce a network that is dimensioned to handle the specified traffic. To the best of our knowledge, our work is also the first one that tries to automatically determine the best network topology while considering network switching costs and propagation delays. We study how the best topology changes on different substrates as traffic conditions vary. In general, we find that as pairwise traffic constraints and delay bounds are relaxed, the leastcost metanet topology becomes increasingly “tree-like”. We also show the impact of delay bounds on the network costs under different traffic conditions. Our algorithm produces metanet configurations that are demonstrably close to the computed lower bound and is fast enough to handle substrate networks of practical size. In a diversified internet, metanetworks (“metanets” for short) share a common substrate and offer value-added services to potentially large numbers of users around the globe. Therefore, configuring low-cost metanets with links having enough bandwidth to accommodate all anticipated user traffic is crucial to the success of the metanets. In this paper, we propose a novel pruning algorithm that configures metanets on any given substrate in a cost-efficient way. In contrast to other testbed configuration systems, we solve the metanet configuration problem from a higher level specification and produce a network that is dimensioned to handle the specified traffic. To the best of our knowledge, our work is also the first one that tries to automatically determine the best network topology while considering network switching costs and propagation delays. We study how the best topology changes on different substrates as traffic conditions vary. In general, we find that as pairwise traffic constraints and delay bounds are relaxed, the leastcost metanet topology becomes increasingly “tree-like”. We also show the impact of delay bounds on the network costs under different traffic conditions. Our algorithm produces metanet configurations that are demonstrably close to the computed lower bound and is fast enough to handle substrate networks of practical size.