Term paper: Randomized Algorithms and Heuristics for Join Ordering

In the relational database setting today, large queries containing many joins are becoming increasingly common. In general the ordering of join-operations is quite sensitive and can have a devastatingly negative effect on the efficiency of the DBMS. Scheufele and Moerkotte proved that join-ordering is NP-complete in the general case [4]. For smaller queries however, less than approximately 10 joins, the optimal join strategy can be found by means of dynamic programming. However, the dynamic programming algorithm, proposed in [5], has a worst case running time of O(2 ) (where N is the number of joins), thus for queries with more than 10 joins, it becomes infeasible. In the literature there are many alternative approaches to the join ordering problem, Steinbrunn et al [6] present a good overview. Approaches such as Iterative Improvement, Simulated Annealing, Genetic Algorithms, Two phase optimization and the probabilistic QuickPick etc all provide efficient alternatives although producing sub-optimal solutions to the join-ordering problem. In the next section a presentation to each of these will be made.