Traffic Splitting and Its Application to Network-Wise Performance Analysis

Traffic in a network may split into several streams when passing through an intermediate node, e.g., router or switch. Then one question may arise: what do the resultant traffic streams become? Trying to answer such question, we characterize the split traffic routed by a probabilistic routing algorithm when the input traffic is modelled by a Markov-modulated Poisson process (MMPP) in this paper. We find out that each split traffic under such splitting mechanism is also an MMPP with a proper parameter modification which is investigated in this paper. After having characterized the split traffic, we apply the result to enable us to get network-wise performance measures, i.e., end-to-end sojourn delay and delay variance in networks employing probabilistic routing algorithms via an approximate approach. With the evidence of numerical experiments, we show that the accuracy for those performance measures remains in an acceptable region. Therefore, such network-wise performance tool may have widespread applications in networks, such as the Internet and asynchronous transfer mode (ATM) network.