Research Of Load-balancing Strategy Based On Content-oriented Switch On Linux Cluster

The current computer technology has entered into the era of network-centered, Web user numbers, application types, and network flow increase exponentially; furthermore, different application has different flow and computing requirements. The method that upgrades single server's computing capabilities only can't meet the humankind's needs more and more; the main problems are presented in the following aspects, such as upgrade process complicacy, high cost and service discontinuity and the like. LVS (Linux Virtual server) cluster technology can meet the network service's scalability, high availability, serviceability, price effectiveness needs. The basic principle to realize the network service scalability for LVS is through the one-to-multiple mapping mechanism. Forwarding the service request to multiple nodes to execute, this mapping can be realized in multiple-layers. Under the framework of LVS,the exchange technology in the 4th layer-IPVS(IP Virtual Server) and KTCPVS (Kernel TCP Virtual Server) in the 7th layer are realized. The exchange in the 7th layer is also called content-oriented exchange technology. It can forward to respective server according to different customer request contents.In the common Web applications, task has priority. The 4th layer oriented exchange can't differentiate the tasks, which requests background servers to provide same service, Regardless of which server the request is being sent to, the return result should be same. But the exchange technology in the 7th layer balances the load, based on the contents. It provides a high-level control method for visiting flow. It executes the load-balancing decision by examining the data message in the application layer, and action according to the information in the message. It also can use the local features of the visiting completely- send similar requests to the identical server, which makes good similarity of the requests, received at each background server. It can increase cache efficiency of each single server further; consequently it improves the system performance under the limited hardware configuration. Accordingly it's of high commercial value in the WEB cluster loading field.The study object of this paper is on KTCPVS of LVS, content-oriented exchange strategy research is studied. The following major aspects are included, the basic principle of content-oriented exchange, realization mechanism. The idea that allocates and differentiates the tasks according to priority is brought forward. This paper makes out the allocation arithmetic of task value ratio maximization when the cluster system loading is heavier. The research of this paper include the following aspects:①It gives a general introduction to development, research status, related issues, and realization principle of loading balancing technology. It expounds the realization of loading balancing, which is the key technology to cluster, using LVS as a sample. System structure of KTCPVS in the 7th layer and allocation arithmetic, based on contents are explained in details.②It expatiates on the principle of content-oriented exchange, analyzes the advantage of content-oriented exchange, compares and summarizes the realization of content-oriented exchange technology, such as application proxy, TCP Gateway, TCP Splicing, TCP Handoff and so on.③It brings forward the adjustment mechanism of dynamic loading realization at Linux kernel and the strategy that allocates tasks according to task value ratio when the cluster is overloaded. The server comprehensive loading is computed by quantifying allocation meter and real time loading information feedback from real servers, consequently measures the comprehensive loading status of the whole cluster accurately. The details of using allocation arithmetic of task value ratio maximization to allocate the loading, while the cluster is overloaded are elucidated.Lastly, the actual testing and real application proves that content-oriented exchange and task value ratio maximization allocation arithmetic in this paper can utilize the computing resources of Linux cluster system at most and satisfactory task-handling effects are achieved.