In peer-to-peer video systems, most hosts will retain only a small portion of a video after its playback. This presents two challenges in managing video data in such systems: (1) how a host can find enough video pieces, which may scatter among the whole system, to assemble a complete video, and (2) given a limited buffer size, what part of a video a host should cache. In this thesis, we address these problems with a new distributive file management technique. In our scheme, we organize hosts into many cells, each of which is a distinct set of hosts which together can supply a video in its entirety. Because each cell is dynamically created and individually managed as an independent video supplier, our technique addresses the two problems, video lookup and caching, simultaneously. First, a client looking for a video can stop its search as soon as it finds a host that caches any part of the video. This dramatically reduces the search scope of a video lookup. Second, caching operations can now be coordinated within each cell to balance data redundancy in the system. We have implemented a Gnutella-like simulation network and use it as a testbed to evaluate the proposed technique. Our extensive study shows convincingly the performance advantage of the new scheme.