Research On Load Balancing System Based On Hot Contents

The parallel database system consists of many servers provides a excellent platform for e-commerce and e-government. The share-nothing database cluster is extremely sensitive of load skew. The load skew will seriously affect the performance of the share-nothing clusters without load balancing mechanism. The current research on such issues is limited to the solution of the load skew caused by only one reason.To cope with all kinds of load skew, first, we divide the reason of load skew into two categories: data skew and execution skew. Then, We analyse how to find out which kind of reason lead to the load skew. After got the reason, system can take appropriate method to change the data distribution to achieve load balancing purposes. During the analysing, the System is always trying to find and move the data which affect the performance most seriously, so we call this load balancing method for load balancing method based on hot contents.This method consists of five algorithms: load monitor algorithm, data skew judgement algorithm, data skew analysing algorithm, data skew processing algorithm and execution processing algorithm. The data skew judgement algorithm and data skew analysing algorithm can find out the reason of load skew. The data skew processing algorithm and execution skew algorithm can solve data skew and execution skew.The most attractive propery of the hot contents based load balancing method is: it can cope with the data skew which is caused by data skew and execution skew. In the process, the method needn't the information about phyiscal storage, so it has execellent adaptability. The experimental results show that the hot contents based load balancing method can cope with all kinds of data skew, and it has good adaptability and utility.