Heuristic Solutions for A Mapping Problem in A TV-Anytime Server Network

An electronic newspaper can be sent to users at anytime, but broadcast of video media is still strictly time dependent. Current ratio of storage capacity to price makes it possible to store a large number of video data of high quality with relatively low price. As the last-mile bandwidth is being solved with the broadband technologies such as ADSL and CATV, the contents sent by current wired TV-net in a broadcasting way may be offered to users interactively and time independently. The service which implement such characterizes is called as TV-Anytime service. In this paper a hierarchical architecture of server network is proposed to support a metropolitan TV-Anytime system. A key issue, video objects mapping problem in such a distributed server network, is studied and formulated as a combinatorial optimization problem. It takes total storage capacity of the server network, communication bandwidth of the backbone networks and access patterns into account. The objective is to provide each requested video stream by use of the highest possible encoding bit-rate. We identify it as the Quality of Service here. A nice feature of media is the fact that a single video stream can be encoded in different bit-rates, providing different presentation quality. It also uses some different storage capacities and communication bandwidths. Thus, the tradeoff that has to be solved by the mapping problem is to provide each client the selected media assets in the highest possible quality, and take the restrictions imposed by the underlying hardware into account. We present a number of heuristic methods based on simulated annealing for the solution of this video object mapping problem. A three-step initial algorithm is proposed to work out an initial solution. A family of two-phase neighborhood structure algorithm is used to find the near to optimal solutions. We use a parallel simulated annealing library(parSA) to test a set of benchmark instances. The performance of presented methods is evaluated by use of the set of benchmark instances. It shows that the solutions based on parSA can get desirable performance and reach an approximately optimal solution in a short time. The performance also reveals that a more effective system could be obtained by redundant connections of server networks and tracing routing.