Network calculus based simulation for TCP congestion control: theorems, implementation and evaluation

In this paper, we examine the feasibility of incorporating network calculus based models in simulating TCP/IP networks. By exploiting network calculus properties, we characterize how TCP congestion control - additive increase and multiplicative decrease (AIMD), together with the queue management strategy used in routers, regulates TCP flows. We first divide the time axis into intervals (each of which consists of multiple round-trip times), and derive a TCP AIMD throughput model which derives the attainable throughput of a TCP flow, given the number of collisions in an interval. Then based on the derived throughput model, we define a set of network calculus based theorems that give upper and lower bounds on the attainable TCP throughput in each interval. Finally, we implement network calculus (NC) based simulation in ns-2, conduct simulation in both the packet mode and the network calculus-based mode, and measure the performance gain (in terms of the execution time thus reduced) and the error discrepancy (in terms of the discrepancy between NC-based simulation results and packet-level simulation results). The simulation results indicate an order of magnitude or more (maximally 30 times) improvement in execution time and the performance improvement becomes more pronounced as the network size increases (in perspective of network capacities and number of flows). The error discrepancy between NC-based simulation and packet-level simulation, on the other hand, is minimally 1-2% and maximally 8-12% in a wide spectrum of network topologies and traffic loads employed in this study. The simulation results also indicate the superiority of NC-based simulation over fluid model based simulation (the latter realized using the time stepped hybrid simulation).