Traffic-aware network update in software-defined NFV networks

Recently, a new paradigm integrating software-defined networking (SDN) with network function virtualization (NFV), named SDNFV, has gained increased interests from industry and academia. In SDNFV, network update is a notable process that replaces the current network configuration by the target one. To accomplish the process, network providers modify a set of rules for the data plane in a manner that directs network traffics to a sequence of virtualized network functions (VNFs) in a predefined order. In contrast to the legacy SDN environment, there exists a unique property for SDNFV that the traffic volume of a flow would be changed after the flow is processed by VNFs. We observe that the traffic-changing property causes a "chain-reaction"phenomenon in which both the link bandwidth and CPU utilization of VNFs are affected. Without careful consideration of the phenomenon, the overall performance of SDNFV will be significantly degraded and some consistency properties cannot be maintained during network updates, which motivates us to have in-depth investigation on the phenomenon and its caused problem. We formulate the problem, based on the Fat-tree architecture which has been widely adopted for large-scale data-center deployment, and propose an efficient algorithm to solve the problem. Simulation results demonstrate that our proposed solution can benefit from the proper treatment of traffic-changing property during network updates for SDNFV.