Research On Autonomous Load Balancing In Cluster Environment

With the rapid rise of cloud computing and mobile Internet, more and more computing tasks and storage resources are placed on the servers. Everywhere in the country is building new data center to deal with the explosion of data. These amazing data is propped up by the hundreds of thousands of servers. These servers work together to form clusters providing external service.Load balancing technology of cluster ensure the maximization of efficiency. However, with the deep application of cluster systems, especially the mature of virtualization technology, more load are coming from the cluster itself, but the traditional load balancing strategy can’t handle this situation.The traditional load balancing strategy focus on the external loads of cluster, these loads have a remarkably characteristic is that they are initiated by external clients, and then directed to a server within the cluster via some sort of load balancing algorithm. For this type of load, the traditional load balancing strategies could handle it well. But for the servers’ internal load, the previous strategy becomes useless.This thesis proposes a novel load balancing strategy- Autonomous Load Balancing, aimed at handling the load from within the cluster. Autonomous load balancing regards physical server as a single node, takes the process on the physical servers as load, gets running states of neighbor nodes through the exchange of information among nodes. When the node exceeds the critical value of the load, it will calculate the most appropriate migration target node, and then execute migration. Autonomous load balancing technology enables each node "realize" their load, "communicate " with other neighbor nodes, and through "negotiation" with other nodes to transfer its’ load, so that each node has the ability to self-management, making the entire cluster is load-balanced state.Firstly, the thesis studies the characteristics and application scenes of traditional load balancing technology, pointing out its deficiencies. Secondly, the thesis studies the core problem of load balancing techniques and propose solutions of autonomous load balancing strategy. Thirdly, from the practical application, the thesis proposes a method for calculating the load capacity providing theoretical support after the sequence of scheduling strategy. Lastly, depending on the actual application scenario, the thesis designs two simple and efficient sorting algorithm to sort the target nodes in migration.On the basis of theoretical research, the thesis also designs a more complete autonomous load balancing system and its implementation. In this thesis, the system in a real test environment has been tested to prove its effectiveness. The system design uses a peer model, with high scalability and flexibility, ideal for use in large-scale cluster.