Load balancing problems play a special role in the operation of parallel and distribution systems. The core technology of load balancing is the task scheduling. In order to achieve equally distribute the load on processors and maximize their utilities while minimizing the total task execution time, this article analyses a large number of data retrieval requests content. It shows different tasks have different resource occupation rate on the server node and the current server load is also different. The demand of different task types on CPU and I/O resources are not same. So, this paper presents a dynamic load balancing algorithm which is based on task classification, and applies it to the cluster system data retrieval.