Professor Shao Ziyu’s research group of SIST made an important advance in intelligent networking, addressing the following challenge: “How to systematically design effective predictive scheduling algorithms for intelligent networks?” Their results, of high importance to recent software-defined networking (SDN) systems, were published in an article entitled “Predictive Switch-Controller Association and Control Devolution for SDN Systems” which was accepted by the IEEE/ACM Transactions on Networking (IEEE/ACM TON).
Prof. Shao’s Group developed general predictive scheduling algorithms, which can be applied to various intelligent network scenarios, including software-defined networking, edge computing, network function virtualization systems and data streaming processing systems. The corresponding design schemes are shown in Figure 1.
For example, in the case of SDN systems, in order to enhance the scalability and reliability of the control plane, existing solutions adopt either multi-controller design with static switch-controller association, or static control devolution by delegating certain request processings back to switches. Such solutions can fall short in the face of temporal variations in request traffic, incurring considerable local computation costs in switches and their communication costs in controllers. So far, it still remains an open problem of how to develop a joint online scheme that conducts dynamic switch-controller association and dynamic control devolution. Furthermore, the fundamental benefits of predictive scheduling to SDN systems still remain unexplored. In this paper, Prof. Shao’s group identified the non-trivial trade-off in such a joint design and formulated a stochastic network optimization problem which aimed to minimize time-averaged total system costs and ensure long-term queue stability. By exploiting the unique problem structure, they devised a predictive online switch-controller association and control devolution scheme, which solves the problem through a series of online distributed decision makings. Theoretical analysis showed that without prediction, this scheme could achieve near-optimal total system costs with tunable trade-off for queue stability. With prediction, this scheme could achieve even better performances with shorter latencies. Notably, with mild-value of future information, this scheme incurred a significant reduction in request latencies, even when faced with prediction errors.
Students from Prof. Shao’s group, Huang Xi and Bian Simeng, are the authors. Prof. Shao Ziyu is the corresponding author. This work was conducted at ShanghaiTech University, and supported by the start-up funding from ShanghaiTech University, the National Natural Science Foundation of China, and the Natural Science Foundation of Shanghai.
Figure 1: Examples of Predictive Scheduling Algorithms
Paper Link: https://ieeexplore.ieee.org/document/9200338