Investigation Of Dynamic Resource Allocation In Inter-Datacenter Networks Over Optical Infrastructure

With the rapid development of cloud computing and e-Science, optical Inter-DC networks have attracted intensive research attentions. Inter-DC networks have huge peak throughput and great burstiness, and requests of Inter-DC networks are band-width hungry. Efficient support from the physical layer for transmission is highly needed. Recent advances in EONs have demonstrated agile spectrum management and sub-wavelength switching granularity in the optical layer. EONs are also well known for tremendous bandwidth resources and stability. Hence, EONs can become a promising transmission infrastructure for Inter-DC networks. In order to better study the anycast RSA problem in optical Inter-DC networks, we investigate the unicast RSA problem in EONs at first.We have employed single path routing and multi-path routing to solve the unicast RSA problem. For single path routing, we propose several dynamic RMSA algorithms to achieve a high throughput in network provisioning. It includes two types of algo-rithms:dynamic online path computation RMSA algorithms and dynamic path-set updates RMSA algorithms. We extend the single path routing algorithms to multi-path routing. We have investigated two types of HSMR schemes, which are named as HSMR-OPC and HSMR-FPS. Our simulation results show that HSMR-OPC can achieve the lowest BBP among all HSMR schemes and two other benchmark algo-rithms. This is the first attempt to employ both online path computation and offline path computation in dynamic RMSA using multipath in EONs.For the anycast problem in Inter-DC networks over elastic optical infrastructure, we investigate offline and online RSA problems. The offline RSA problem is a net-work planning problem, and the online RSA problem is a network provisioning prob-lem. We use an ILP model to formulate the offline problem. However, the ILP model is only suitable for small-scale problem because of its high computing complexity. Then we propose several heuristic algorithms which are based on single-DC destina-tion selection strategy to solve the offline problem. We compare gaps of all heuristic algorithms with the optimal results which are provided by the ILP model. For the on-line scenario, we extend the proposed heuristic algorithms in the offline scenario to jointly consider both computing and bandwidth resources for efficient service provi-sioning. For further investigation, we also design a multi-DC destination selection strategy. Our simulation results demonstrate that the best single-DC strategy can re-duce BBP by one magnitude when the traffic load is light, compared with the bench-mark algorithm. While the multi-DC strategy reduce BBP by one magnitude, com-pared with the best single-DC strategy. In short, when we use the multi-DC destina-tion selection strategy to solve the online anycast problem in optical Inter-DC net-works, it can reduce the BBP effectively. The bandwidth resources in elastic optical Inter-DC networks can be fully utilized, and it also makes computing resources be-come the bottleneck in dynamic provisioning. Note that it is much easier and cheaper to upgrade the DC resources rather than optical infrastructure in optical Inter-DC networks.After the investigation of the anycast problem in Inter-DC networks built over elastic optical infrastructure, we investigate the bandwidth resources and computing resources DF problem in such networks as a further study. We first explain the DF procedure in optical Inter-DC networks, and the reconfiguration method is used for DF. Then we design an anycast based DF-RSA algorithm to solve the DF problem. We propose two requests selection strategies and three approaches for choosing destina-tion DCs for selected requests. Simulation results show that compared with no DF scheme, the DF-RSA algorithm can reduce the BBP significantly.