Internet Traffic Generation for Large Simulations Scenarios with a Target Traffic Matrix

The simulation of realistic networks is an important aspect in the development of new protocols or algorithms for communication networks. The traditional approach for such a simulation is to connect a certain amount of sources to network nodes and to measure the link load, end-to-end delay etc. in the simulation in order to evaluate the performance of the network. Network providers have a different view: the traffic intensities in their networks are known, but their target is to optimize the current network. Therefore the first step is to configure the simulation model to generate the same traffic that was measured in the real network. We prescribe traffic intensities via a traffic matrix and propose a method for the allocation of sources for all flows that minimizes the difference between prescribed and measured traffic matrix. We use HTTP/TCP source models for traffic generation since current Internet traffic is mostly TCP traffic. The reactivity of TCP makes it hard to estimate the average rate of a single source, especially since TCP operates not in steady state here due to the small average of the HTTP transfer volumes in the Internet. Nevertheless we derive a simple estimate on the average rate per HTTP/TCP source in the paper that only uses network and protocol properties that are known before carrying out the simulation with the actual parameters. The performance of the proposed algorithm for the allocation of sources was evaluated by simulating three network topologies with various parameters. We show in this paper that the performance of the method is good under various conditions. Key-Words: HTTP, TCP/IP, Simulation, Traffic Generation