We study integrated prefetching and caching in single and parallel disk systems. A recent approach used linear programming to solve the problem. We show that integrated prefetching and caching can also be formulated as a min-cost multicommodity flow problem and, exploiting special properties of our network, can be solved using combinatorial techniques. Moreover, for parallel disk systems, we develop improved approximation algorithms, trading performance guarantee for running time. If the number of disks is constant, we achieve a 2-approximation.