Necessary to be known by this variation of our method. We believe that the assumption on a skewed distribution of movie requests is a pragmatic one. The model considered here follows [AGH, 95], as far as the architecture is concerned. Assume, that the distribution p() of movie requests is known. Our scheduler S divides (initially) the channels according to p(), so that channels in partition C i will be used for movies to beseenby at least i users on average. The goal is to allocate channels in a way which optimizes channel revenue and reduces unused channels. S puts the arriving requests to the movie queues: When a request for movie m j is done, it is inserted into Q j . When 3 Work supported in part by the ESPRIT II Basic Research project GEPPCOM under contract No. 7141, and by the IT Long Term Research project ALCOM-IT under contract No.