Joint server and route selection in SDN networks

The key computing trends driving the need for a new networking paradigm include evolving traffic characteristics with the rise of mobile and cloud services, network function virtualization, the need for more capacity, and the trend of building networks without vendor dependence. The SDN architecture, by decoupling data and control planes and using a centralized controller, provides network managers with the flexibility of configuring, managing, and securing network resources dynamically and in an automated fashion. With the support of SDN tools, network resources can be utilized efficiently, evolving business needs can be addressed timely, and higher transmission capacity can be achieved. The current operational networks do not have enough flexibility in terms of managing how available network resources can be efficiently utilized since data and control plane functions are coupled and networking functions are highly distributed. A traditional network management tool might be considered as a central controller; however, its operations would be restricted by the networking protocols of the underlying OSI layers. In the SDN architecture, data forwarding and control planes are decoupled, and hence, it becomes feasible to utilize logically centralized controller for intelligent network management decisions. This approach is directly applicable for small size regional networks, data center networking, and the core network of a cellular wireless communication system. In the SDN architecture, SDN enabled switches, which can be managed by a centralized controller, are used instead of standard network equipment. SDN networks are vendor independent since information exchange between switches and the controller is provided by a standardized interface. OpenFlow is the first standard communication interface defined between control and forwarding layers of SDN. OpenFlow allows direct access to SDN enabled switches for information retrieval and configuration purposes. The SDN controller software can manage the underlying network either statically or dynamically and uses the flow concept of OpenFlow to identify network traffic based on matching rules. Managing the network on a per flow basis provides a fine granular resource control and timely response to network changes. Intelligently routing flows using a centralized controller with network-wide state information is expected to provide significant performance improvements in terms of throughput, delay, and packet loss. These improvements are particularly important to support new real-time services with stringent quality of service requirements. However, the timely detection of congestion and failure events and re-routing traffic without causing any oscillation are challenging tasks. The SDN technology with a centralized controller is suitable for tackling these problems. Although a programmable centralized controller gives flexibility in flow management; it is a challenging task when the network size is large and there are many flows with various priority levels. In the traditional SDN approach, the forwarding decisions are performed using the traditional shortest path algorithm running in the SDN controller. Therefore, the packets are forwarded over the shortest path by data plane elements (i.e., OpenFlow switches) without considering link utilization levels. This situation may cause bottlenecks on certain links while there might be under-utilized links in other parts of the network. In this thesis, a joint server and route selection mechanism is presented for SDN networks. OpenFlow enabled architecture takes current and past network statistics into account and routes an incoming flow over under-utilized server and links to evenly distribute traffic among the network resources. The main objective is to use the current and past knowledge of the network so that a better resource allocation decision can be made by considering both current and future times. The proposed algorithm uses a cost function to periodically and automatically modify link weights of link state routing protocols using current and past network statistics such as traffic and packet loss rates. There are functionally equivalent multiple servers in the network and the server utilization statistics are also used in the overall cost calculation for joint server and route selection. When a new traffic flow is arrived, the proposed algorithm intelligently selects the server and routing path by using the overall cost calculation of the algorithm. Finally, new routes corresponding to the minimum cost are installed into OpenFlow enabled switches. New SDN applications are developed to implement the proposed traffic engineering (TE) mechanism. These applications communicate with the network using the controller (controller APIs and network service functions) for collecting statistics and re-configuring switches. Custom database (DB) is used to store collected information. The proof of concept has been implemented using Mininet to emulate OpenFlow enabled switches and OpenDaylight as an SDN controller. The demonstration results indicate that the proposed mechanism can evenly distribute traffic among the links and servers and avoid congestion in the network.