Abstract: In the context-aware application server, large amount of context information must be collected and processed. Server cluster and load balance are critical technique to ensure the availability and reliability of context-aware server. This paper presents the predicting based load balance algorithm, which allocates task dynamically using SVM to predict the system load in next time phase and consider the current load of all nodes in the system at the same time. The task cannot be transferred to another machine when it is already allocated to one node, which can reduce the consumption of system resource. Comparison of load balancing algorithms shows that the prediction algorithms can improve the performance of server when the system load changes dramatically.

Key words: dynamic load balance, prediction algorithms, SVM, context-aware

摘要： 在上下文感知服务器中，需要收集大量的上下文信息并对其进行处理。必须通过服务器集群并采用负载均衡技术才能保证服务器的正常运行。该文提出基于预测负载进行自适应负载均衡的算法，采用支持向量机预测系统运行下一个阶段的负载量，结合当前系统各节点处理情况进行动态负载分配过程，在任务执行过程中，不进行任务的迁移以减少迁移所带来的系统消耗。通过比较表明，在负载变化较大的情况下，算法能够有效地提高服务器性能。

关键词: 动态负载均衡, 预测算法, 支持向量机, 上下文感知

CLC Number: 

• TP393