Near-optimal data allocation over multiple broadcast channels

AbsractData broadcast is known as a scalable way to transmit database items to large client populations through a wireless channel. However, with a large set of items, the expected delay of receiving desired data increases due to the sequential nature of the broadcast channel. One possible solution is to increase the number of available channels and allocate items evenly over these channels, then cyclically broadcast data in each channel. In view of data access skew (the access frequencies of data are usually different), some channels may be reserved exclusively for those few frequently requested items, while the infrequently accessed bulk of data are allocated on other channels. In this paper, an O(N log K) restricted dynamic programming (RDP) algorithm is proposed to partition N items over K channels assuming data access is skewed with the object of minimizing the average expected delay (aed) of clients. To speed up the DP process, for any partition, we predict a possible location which may be very close to the optimal cut by using a low bound as the actual aed for the remaining items. Thus, the number of comparisons in DP can be restricted to a small interval around the predicted cut point. To further reduce the costs in DP, the hierarchical property of optimal solution is adopted. Simulation results show that, the hit rate obtained by RDP is higher than 90% and it also outperforms the existing algorithm 200%. We extend the work of RDP and develop an O(N log N log K) PKR algorithm; simulation results show that the solution is optimal.