Proxylib servers have been successfully used in reducing playback latency and network bandwidth required for delivering VBR videos. By storing portions of video programs in local video proxies and the rests in remote servers, the peak and variability of network bandwidth requirements can be reduced. Regarding to which portion of video frames should be cached at higher priority, let's consider an MPEG video in which an I-frame is referenced by all the other frames in the same GOP. Obviously, losing a packet belonging to an I-frame makes it difficult to recover all the frames of the same GOP. In this paper, based on the video proxy problem introduced by [1], we present the Optimal Caching (OC) algorithm to minimize cache size required for delivering a specific video program subject to network bandwidth and client buffer size constraints, and the Optimal Selective Caching (OSC) algorithm to further select the most appropriate portions of a video program for caching in the video proxy. In our experiments, it is shown that our algorithm is more effective than the previous algorithm. Our OC and OSC algorithms reduce cache size by 50% as compared with the algorithm presented in [1]. Furthermore, the ratio of I-frames cached in the video proxy is 5 times larger than the result of applying the algorithm presented by [1]

¡@