An experimental evaluation and guideline for path finding in weighted dynamic network

Shortest path computation is a building block of various network applications. Since real-life networks evolve as time passes, the Dynamic Shortest Path (DSP) problem has drawn lots of attention in recent years. However, as DSP has many factors related to network topology, update patterns, and query characteristics, existing works only test their algorithms on limited situations without sufficient comparisons with other approaches. Thus, it is still hard to choose the most suitable method in practice. To this end, we first identify the determinant dimensions and constraint dimensions of the DSP problem and create a complete problem space to cover all possible situations. Then we evaluate the state-of-the-art DSP methods under the same implementation standard and test them systematically under a set of synthetic dynamic networks. Furthermore, we propose the concept of dynamic degree to classify the dynamic environments and use throughput to evaluate their performance. These results can serve as a guideline to find the best solution for each situation during system implementation and also identify research opportunities. Finally, we validate our findings on real-life dynamic networks.

Path computation enhancement in SDN networks

Path computation is always the core topic in networking. The target of the path computation is to choose an appropriate path for the traffic flow. With the emergence of Software-defined networking (SDN), path computation moves from the distributed network nodes to a centralized controller. In this thesis, we will present a load balancing algorithm in SDN framework for popular data center networks and a fault management approach for hybrid SDN networks. The proposed load balancing algorithm computes and selects appropriate paths based on characteristics of data center networks and congestion status. In addition, a solution that supports proper operations of a hybrid SDN network will also be proposed. The evaluation shows the proposed load balancing algorithm performs better than classic shortest path algorithms. We also demonstrated that the proposed solution for hybrid SDN networks can support proper operations in complicated hybrid SDN networks.