Network Invariant-Based Fast Simulation for Large Scale TCP/IP Networks

In this paper, we present a rescaling simulation methodology (RSM) to expedite simulation in large-scale TCP/IP networks without loss of fidelity of simulation results. Conceptually, we scale down the network to be simulated to reduce the number of events, simulate the downscaled network for a short period of time, and then extrapolate the corresponding results for the original network by scaling up the simulation results obtained from the downscaled network. Both the operations of scaling down and rescaling up the network are conducted in such a manner that the network invariant, called the bandwidth-delay product, is preserved. In particular, since the dynamics of queues, such as the queue size, the dropping probability, and other parameters at every link are the same in both the original and downscaled network, RSM can accurately infer the network dynamics behavior of the original network (equipped with various Active Queue Management (AQM) strategies). In contrast to SHRiNK, RSM does not make any assumption on the input traffic and can work with any AQM strategy. It also preserves the queue dynamics and the network capacity as perceived by TCP connections. To validate the proposed methodology, we have implemented RSM based simulation in ns-2, and conducted a simulation study comparing RSM based simulation against packet level simulation, with respect to the capability of capturing transient, packet-level network dynamics, the execution time and the discrepancy in simulation results. The simulation results indicate an order of magnitude or more improvement (maximally 50 times) in execution time and the performance improvement becomes more prominent as the network size increases (in terms of number of nodes and network capacity) or as the scaling parameter decreases. The error discrepancy between RSM based simulation and packet level simulation, on the contrary, is minimally 1-2 % and maximally 10 % in a wide variety of network topologies (with various AQM strategies) and traffic loads. The encouraging simulation results, coupled with the fact that implementation of RSM is simple and straightforward, suggest that RSM can be used to simulate, and accurately infer network dynamics of, large-scale TCP/IP networks