Improving disk efficiency in video servers by random redundant storage

Random redundant storage strategies have proven to be an interesting solution for the problem of storing data in a video server. Several papers describe how a good load bal ance is obtained by using the freedom of choice for the data blocks that are stored more than once. We improve on these results by exploiting the multi-zone character of hard disks. In our model of the load balancing problem we incorporate the actual transfer times of the blocks, depending on the zones in which the blocks are stored. We give an MILP model of the load balancing problem which we use to de rive a number of good load balancing algorithms. We show that, by using these algorithms, the amount of data that is read from the fast zones is substantially larger than with conventional strategies, such that the disks are used more efficiently