Mining heuristics for scheduling distributed hard real-time tasks

We present a method for producing heuristics to direct the search for solutions in task allocation and scheduling problems. The problems considered consist of allocating and scheduling tasks with precedence and hard real-time constraints into distributed systems. Heuristics are produced in two steps. In the first example of promising and unpromising search, states are extracted from the search trees of previously solved problems. In the second, using the extracted examples, the C4.5 algorithm induces classifiers that label search nodes as promising or unpromising. We conclude discussing the results of experiments that compare the success ratio and number of nodes visited to solve problems when a search algorithm chooses the next node to be examined using a random heuristic and when it uses the induced classifiers.