Realistic Large-Scale Online Network Simulation

摘要: Large-scale network simulation is an important technique for studying the dynamic behavior of networks, protocols, and emerging classes distributed application (e.g. Grid, peer-to-peer, etc.) realism are two critical requirements simulations Grid studies. Our work here extends previous efforts in three key ways. First, we study networks 100x larger than our studies (20,000 routers). Second, at this scale, realistic struct ures (100 AS’s, BGP4 OSPF routing) versus flat routing. Finally, describe evaluate a new profile-based load-balancing approach called hierarchical load balance. extensive large-scale experiments with balance (PROF) on flat-routed (OSPF) show that PROF outperforms several other techniques based topology static information. However, these results those multi-AS motivate invention (HPROF) which clusters nodes to achieve desired minimum link latency (MLL), determinant parallelism, then applies graph partitioner. HPROF explicitly controls tradeoff between efficiency available producing robust superior performance including both single-AS networks. can improve imbalance by 40%, reduce time about 50% 20,000 router executed 128-node clusters. The parallel achieved over providing substantial capabilities simulating large In summary, advances demonstrate routers (comparable Tier-1 ISP like AT&T) be accomplished system.