Cutting Long-Tail Latency of Routing Response in Software Defined Networks.
Junjie Xie,Deke Guo,Xiaozhou Li,Yulong Shen,Xiaohong Jiang
DOI: https://doi.org/10.1109/jsac.2018.2815358
IF: 16.4
2018-01-01
IEEE Journal on Selected Areas in Communications
Abstract:To enable the network softwarization, network function virtualization (NFV) and software defined networking (SDN) are integrated to jointly manage and utilize the network resource and virtualized network functions (VNFs). For a network flow resulting from any NFV application, an associated switch would send a routing request to the controller in SDN. The controller then generates and configures a routing path to dynamically steer the flow across appropriate VNFs or service function chains. This process, however, exhibits a skew distribution of response latency with a long tail. Cutting the long-tail latency of response is critical to enable the network softwarization, yet difficult to achieve due to many factors, such as the limited capacities and the load imbalance among controllers. In this paper, we reveal that such flow requests still experience the long-tail response latency, even using the up-to-date controller-to-switch assignment mechanism. To tackle this essential problem, we first propose a light-weight and load-aware switch-to-controller selection scheme to cut the long-tail response latency under the simple scenario of homogeneous controllers, and then design a general delay-aware switch-to-controller selection scheme to fundamentally cut the long-tail response latency for the more complicated heterogeneous controller scenario with performance fluctuations. The comprehensive evaluations indicate that our two new switch-to-controller selection schemes can significantly reduce the long-tail latency and provide higher system throughput.