Modeling and optimization of the data plane in the SDN-based DCN by queuing theory

Complicated management tasks and various functions of traditional Data Center Network (DCNs) motivate the research on Software-Defined Networking (SDN). However, the limited flow table space in the SDN switch has a significant impact on the data plane especially the availability. In this paper, we propose a theoretical model based on queuing theory to estimate the flow table states of SDN switches, providing operators insights into the requirements of flow table resources and guaranteeing the performance of the data plane. First, we analyze the lifecycle of a flow table entry and construct a queuing theory based estimation model to estimate flow table states in the data plane. Second, we observe the real workloads of DCNs and figure out the probability characteristics of routing strategies to calculate the key parameters in our model. Third, for achieving an optimally latency-aware data plane, we propose an auxiliary method, Amora, to maintain the availability of the data plane by improving routing decisions with the consideration of the optimal flow table states obtained by our model. Comprehensive experiments show that the relative estimation error of flow table states in our model is around 10%, which exposes the actual flow table states and benefits the selection of proper switches. Moreover, Amora efficiently eliminates the failure cases of establishing routing paths in the data plane, which further demonstrates the precision and practicability of our estimation model.