Traffic And Resource Management For Next Generation Networks

Cloud computing and Software-Defined Networking (SDN), the core technologies for thenext generation networks, are hot research topics in the field of computer networking. Cloudcomputing is a model for enabling ubiquitous, convenient, on-demand network access to ashared pool of configurable computing resources (e.g., networks, servers, storage,applications, and services) that can be rapidly provisioned and released with minimalmanagement effort or service provider interaction. Cloud service providers use data centermanagement, large-scale data processing, application deployment, and other technologies toachieve the service of computer resources; users can quickly apply and release resource basedon their demand and pay for what they use, which improves the efficiency of service andreduces users’ operational cost. SDN decouples the function of forwarding decision in controlplane logic from the function of packet forwarding in data plane in networking devices (e.g.,switches or routers). Forwarding decisions is achieved by programmable software whilepacket forwarding is implemented by simple hardware, which simplifies networkmanagement and configuration operation and enables flexible traffic control and thedeployment of innovative networking applications.In order to popularize the next generation networks, the thesis studies traffic and resourcemanagement for cloud computing and SDN. Traffic and resource management mechanismsproposed in the thesis promote the development of the next generation networks by reducingthe electricity cost of data centers, lowering the controller’s workload in SDN, and achievingresource balancing in the data plan of SDN.The main contribution and innovations of the thesis are listed below:(1) We propose AggreFlow, a traffic management mechanism for green data center network.AggreFlow is based on the concept of flow-set. The controller’s routing overhead is reducedby using flow-set routing, and the control message storm is prevented by using just-in-timeflow-set rerouting. In addition, AggreFlow adaptively reroutes some flow-sets to improve theperformance of load balancing in the network. AggreFlow solves the problem of unbalancedload, high routing and rerouting overhead caused by current traffic management mechanismfor green data center network.(2) We propose Joint ElectriciTy price-aware and cooling efficiency-aware load balancing(JET), a resource management mechanism to reduce the electricity cost of distributed datacenters. By jointly considering time-varying locational electricity prices and the impact ofworkload distribution in a datacenter on the efficiency of a cooling system, JET alternately selects the dominating factor of the total electricity cost based on the number of servicerequests and the operational status of distributed data centers. Thus, JET achieves thetrade-off between the electricity cost of active servers and cooling systems and significantlyreduces the total electricity cost of distributed data centers.(3) We propose Load Variance-based Synchronization (LVS), a traffic managementmechanism for the multi-domain multi-controller SDNs. LVS conducts state synchronizationsamong controllers only when the load of a specific server or domain exceeds a certainthreshold. LVS eliminates forwarding loops by enabling controllers to have the consistentinformation about the least loaded server or domain, and significantly reduces thesynchronization overhead of controllers by lowering synchronization frequency.(4) We propose Software-defined trAffIc and fLow-table aware routing (SAIL), a resourcemanagement mechanism for SDNs. SAIL conducts adaptive routing for new flows based ontime-varying link and flow-table utilization to achieve Joint Traffic and Flow-tableEngineering (JTFE) that enables joint traffic and flow-table balancing for SDNs. SAILreduces the controller’s workload and flow completion time, and improves networkperformance.