Abstract: A Web cache-cluster system with higher throughput and hit rate was proposed. The mechanisms of a single cache map and load balance were introduced in the system. Request was regarded as granularity to calculate load. The process power of cache-node was studied from four aspects: network bandwidth, memory capacity, disk access rate, and CPU usage. And the weighted load of cache-node was customized. Based on this, a load-balancing algorithm was proposed. Polygraph was used as a benchmark tool to test our cache cluster and the cache cluster with CARP protocol respectively. Experiment results show that the performance of our cache cluster system is much better.