Reading a Set of Disk Pages

The problem studied in this paper is as follows Consider a le stored in contiguous space on disk Given a list of pages to be retrieved from the le what is the fastest way of retrieving them It is assumed that adjacent pages on disk can be read with a single read request The straight forward solution is to read the desired pages one by one However if two or more pages are located close to each other it may be faster to read them with a single read request possibly even reading some intervening empty pages It is shown that nding an optimal read sched ule is equivalent to nding the shortest path in a certain graph A very simple approximate algorithm is then intro duced and experimentally shown to produce schedules that are close to optimal The expected cost of schedules produced by this algorithm is derived It is found that signi cant speed up can be achieved by the simple mech anism of using additional bu er space and issuing large reads whenever it is advantageous to do so