Load Balancing Strategy Based On Software Aging

With the rapid development of computer network and technology, application server cluster has been increasingly more widely used. Among this, load balancing strategy has mostly decided whether a server cluster can provide efficient and robust service. Load balancing strategy is a mechanism which using some certain distribution strategy to balance the workload on each node, in order to make them balance with each other. However, with the nodes running continuously, some phenomenon, such as exhaustion of resource, data corruption and errors accumulation, may probably happen. With the time elapsing, capacities of nodes may degrade and even cause the nodes to be in failure. Such phenomenon is called software aging in the server nodes.This paper is using new load balancing strategy based on software aging, and change the traditional uniform load balancing strategy to non-uniform strategy. In the software aging prediction, Naive Bayes classifier based on Markov Chain is used. Load balancing algorithm is to plug into software parameter based on original algorithm. When predicting the aging degree and trend, this paper uses several aging metrics as the input and deploys the Naive Bayes classifier based on Markov Chain. First, we obtain the different feature trend from the Naive Bayes classifier. Second, capture the transform of feature distribution while using the Markov Chain method. Finally, classify the point in the aging feature space through Naive Bayes classifier. When designing the non-uniform load balancing strategy based on software aging, plug the software parameter into the original load balancing algorithm.The experiments show that the prediction accuracy of aging prediction based on several metrics may be more than that based on a single metric. Moreover, load balancing algorithm based on software parameter has more advantage than the original algorithm without software parameter.