Abstract: Buffer management is a very critical problem in large-scale video streaming servers. Especially with the appearance of IA-64, the physical memory size has increased up to 16 exabytes, so the buffer management becomes more and more important. There are already many caching policies, among which interval caching policy has been proposed as an effective one. But most of the previous interval-based policies don't consider the popularity of video objects, and the use of huge memories provided by IA-64 systems. The memory utilization is troubled by this. A popularity-based interval caching policy (PIC) is presented to solve the problem. It makes use of the huge memory in IA-64 systems and pays attention to the popularity of video objects. To study the performance of this policy, a static analytical model is given, and large amount of simulations are conducted. The results show that the PIC policy outperforms the traditional interval caching policy.