Combating I-O bottleneck using prefetching: model, algorithms, and ramifications