Synergistic Resource Scheduling In Optically Interconnected Cloud Data Center

Recently, with the rapid development of cloud computing services, Data Center (DC), as the core of Cloud infrastructure, is playing a more and more important role. Tremendous applications run on the DCs, leading to the exponential growth of data volume, which in turn requires higher performance of intra-DC networks in terms of bandwidth, energy consumption, latency, flexibility and reliability. Traditional electrical switching technology might encounter bottlenecks and those critical requirements cannot be satisfied. Optical switching is regarded as a promising technology in DC networks with the advantage of high bandwidth, low latency and better energy efficiency. Meanwhile, considering that there are a messy of coexisted computing, storage and network resources inside DC, the collaborative orchestration of these resources can be vital to improving quality of services for users, and also a great challenge for current DC operation. In light of these, this thesis studies the coordinated control schemes in optically interconnected cloud DC, providing solutions to achieve synergistic resource scheduling by using centralized control plane technologies.Firstly, based on the Optical Burst Ring (OBRing) Network, we propose a centralized control scheme featuring the ability of exploiting all-optical multicast transmission. The workflow of multicast request assignment and timeslot mapping is described in detail, and a corresponding extended OpenFlow-based control plane is proposed. With the interaction of OpenFlow controller and OpenStack orchestrator, we realized the collaborative orchestration of compute, storage and optical network resources. Taking an example of virtual machine deployment, we further illustrated the advantages of the proposed schemes in that the target location of the virtual machine can be chosen with joint consideration of both computing and optical bandwidth resource status and optically multicasting virtual machine images can improve transmission efficiency. Simulation results show that the virtual machine deployment delay can be reduced significantly.Secondly, in order to accommodate inter-DC traffic more efficiently, we also studied the resource scheduling mechanism in IP＋Optical cross-layer networks for the inter-DC scenario. This includes:1) the framework of cross-layer control and optical resource oriented OpenFlow protocol extensions,2) the design and implementation of a unified controller (UC) which supports dynamic path setup, dynamic bandwidth adjustment and dynamic cross-layer restoration,3) experimental demonstrations that verify the effectiveness of our proposals.