Distributing videos to Internet users in real time will always be a challenging problem. A client-server-based architecture faces the predicament of server capability when handling a flux of requests. Therefore, peer-to-peer (P2P) systems are widely adopted for streaming videos to users.

In P2P, each user acts as both a server and a client. In the server role, a user stores parts of the frames in a cache that can be delivered to other users upon request. Because of the limited cache size, a replacement takes place when the cache is full and new frames need to come in. Traditional memory replacement strategies--such as last recently used (LRU) and first in, first out (FIFO)--can always be used. Cheng, Wu, and Lee propose in this paper a modified method, conditioned LRU (C-LRU), to handle the cache replacement.

C-LRU considers two factors. One is the cost of a frame defined as the hop distance. If a frame can only be obtained from a faraway peer, its cost will be high. The other factor is the importance of a frame in reconstructing the video.

The idea is simple. However, its rigorousness is not well demonstrated, perhaps because this is a conference paper. In addition, the English used is unacceptable.