Load balancing approaches for heterogeneous platforms

From the perspective of load balancing, the distribution of flows is analyzed in the following proposals.The author in citep{wang2011openflow} proposes a method for load balancing in which the problem of the source addresses of each client plays a huge problem in the number of “microflows” that are going to be inserted into the switches and the high processing load for the controller. As a solution to this problem is proposed a wildcard computation targeting traffic distribution adjusting automatically flows inserted without disrupting existing connections.In citep{li2013openflow} is proposed a load balancer OpenFlow-based for a datacenter environment, which implements a dynamic routing algorithm for the load balancing decision. It uses a dynamic load balancing algorithm that through the reception of packets in the controller a load balancing function is activated which likewise, enables a recursive search for paths for scalability performance allowing that each alternative path handles equal amounts of traffic load. The searching capability is based on the hierarchical feature of fat-tree networks. Thus, intelligent decision making for load balancing is achieved through the information given by the real-time traffic statistics obtained via OpenFlow.In citep{zhang2011resilience} the author researches the node deployment of controllers from the placement viewpoint. It evaluates the location of the controller nodes from of the perspective of the resilience of the network. It introduces a mathematical approach in which an optimization problem of maximizing the reliability or minimizing the failure probability for the controller placement is proposed. To solve the problem, the author employs graphical partition techniques. It makes use of an algorithm based on the min-cut server placement, which is modified to fit the requirements of the problem. It finds an improvement of the 4\% compared to a random placement and 2\% to a greedy-based placement when the failure probability in the network is equal to 0.01.  In citep{tootoonchian2010hyperflow} Hyperflow was introduced. It presents a solution for control plane platforms distribution, keeping the logical intelligence centralized. HyperFlow aims to minimize the control plane response time to data plane elements requests using a passive synchronization of all OpenFlow controllers. It is done through the implementation of a publish/subscribe system (WheelFS). There are three channels that must be subscribed by the controllers: the data channel, the control channel, and its own channel. The central application publishes network and application events in the data channel. The control channel is used to publish specific commands pointed to a defined controller. The own channel provides a periodical advertisement for discovery and failure detection. Thus, its objective is to provide resiliency to network partitioning and failures.In the environment of NFV and considering the functional tasks exposed in the previous the chapter for a BRAS system as a virtualized network function, the table
ef{tab:Comparison} provides a comparison for each of the approaches presented before.

Leave a Reply

Your email address will not be published. Required fields are marked *