Related work on CPP

Related Work on CPP

[Long Yao et al. 2015] Controller Placement and Flow based Dynamic Management Problem towards SDN

Software Defined Networking(SDN) is a new emerging technology that decouples the control plane from the data plane. It relies on a centralized controller that runs on control plane to manage the network. On one hand, the control plane communicates with the forwarding layer to collect the infor- mation and maintains the network topology; On the other hand, it interacts with the application layer to implement various network functions. Despite the technical benefits SDN brings, especially in the field of traffic engineering [1]. However, the separation of control plane and forwarding plane also brings some new problems, such as the controller placement problem [2] and network management problem under flow dynamics [3] .

The controller placement problem focuses on seeking the best controller position to satisfy the optimal target for a give SDN topology. When there is a single domain, a single controller is needed. While for large scale networks, multiple domain is created and multiple controllers are necessary, thus how to deploy multiple controllers is a research point in SDN. For the controller placement problem, it can be regarded as the task in network designing period which relies on the physical network topology to get the best controller position without considering the network dynamics. In current researches, most of the researchers take the propagation latency between switch- es and controller into consideration, and ignore the weight of switches which is a critical factor in real networks. From this point, we propose a new controller placement metric considering the switch weight and the delay from switches to controller together. The objective of the metric is to minimize the total cost of flow set-up request from switches to controller.

Another problem is the dynamic network management prob- lem, which mainly deals with the flow variations in multiple control domains for large scale networks. The centralized management of SDN confronts the problems of scalability and reliability [4]. Therefore, distributed control architecture is proposed to solve the problem, such as Onix [5] , Hyperflow [6], Devoflow [7], and so on. However, these proposals have the limitation that the mapping between the switches and controllers are static, making it difficult for controllers to adapt to the flow variations. When the switches observe a large number of busty flows, the controller that manages the switches may become overload. Thus, the static mapping may not able to deal with the dynamic flows, and it can easily cause load unbalance of controllers. To address the problem, we propose a flow based dynamic switch migration algorithm to realize the load balance among controllers. When controller becomes overload, it cooperates with the neighbors to migrate the boundary switches to the lightly-load neighbor controllers.

The controller placement problem was first proposed in [2], where the authors considered the best controller positions from the view of minimized average latency and minimized maximum latency from switches to controllers. However, the author didn’t consider the weight of nodes and regard them as the same. In [8] [9] [10] [11], the authors defined the optimal controller placement problem from the network resilience, both the control plane and data plane. We do approval the view of resilience, however, the disruption possibility in network is rare and we may not take this into consideration in the pre-planning of network. In [12], the authors proposed a zone-based distributed network optimization and gave several heuristic solutions. The authors in [13] considered the con- troller capacity in the placement problem. And in [14] the au- thors gave a complete modulation for the controller placement problem from the view of deployment price. However, the proposals mentioned before are all static deployment without considering the flow dynamics in real network.

Recently, Bari in [15] provided a dynamic controller pro- vision strategy based on the flow dynamics. They formulated the optimal controller provisioning problem as an integer linear problem and proposed two heuristic algorithms. And in [3], which was most related with our works, the author proposed the detailed migration mechanism when dynamically assigning switches to controller. However, the author did not give any algorithms to determine the reassigned switches. In recent researches, the authors gave solution from game theory [16] [17] maybe a good solution towards the dynamic management in SDN.

1. Controller Placement for a single domain: In a sin- gle domain, we suppose that the network is managed by a centralized controller. For any given network topology, it is constituted by nodes and links. Therefore, these two issues should be considered simultaneously in the optimal controller placement problem. However, the previous works in controller placement problem only considered the latency from switch to controller [2][8][9][10]. This means that all nodes are considered to be equal in SDN networks, and this is obviously unreasonable.

In SDN, the main function of controller is dealing with the new flow set-up requests. When a packet arrives at the switch, it will match against the flow entries. In the situation of a matched flow entry, the corresponding action is taken. Whereas if there is no matching entries in switch, the Packet-in message will be send to controller, then the controller will determine the rule to handle the packets. In the controller placement strategy, we mainly focus on the cost of flow set-up requests (Packet-in messages) from switches to controller. It consists of the following two issues:

1. •        The number of Packet-in message. •        The routing cost from the switches to controller.

Multiple Controller Placement problem: A centralized controller of SDN confronts the scalability and reliability problem. Therefore, researchers have proposed distributed multiple controllers to realize logical centralized and physical distributed control towards the SDN network. In the multiple controller placement problem, we propose a near-optimum method which refers to the idea of diver-and-conquer. Firstly, the network topology is divided into several sub-domains with topology partition algorithm such as the Multilevel k- way partition[20]. Afterwards, it is demanded to deploy a single controller in each domain, which is the work we have done before. Finally, the controller maintains the mapping relationship with its corresponding switches.

...