Excel Solvers for the Traveling Salesman Problem

Ordering queries within a workload and ordering joins in a query are important problems in databases [1]. We give algorithms for the query sequencing problem that scale (small space) and are efficient (low runtime) as compared to earlier work [4]. The errors are small in practice and we are able to further reduce them using geometric repair. We provide a computational comparison of TSP solvers and show extensive testing on benchmark datasets [25] observing its connection to these ordering problems. 1. PROBLEM STATEMENT Database systems are facing an ever increasing demand for high performance. Either as standalone Oracle, SQLServer or DB2 installations or as a backend to Peoplesoft, SAP or Siebel workloads they are required to execute a batch of queries that contain several common subexpressions. Traditionally, query optimizers like [37], [36] optimize queries one at a time and do not identify any commonalities in queries, resulting in repeated computations. As observed in [3, 39] exploiting common results, multi-query optimization (MQO), can lead to significant performance gains – this requires the queries to be ordered in the workload for memory reuse and reduced disk need. Motivated by the importance for ordering problems, we study the combinatorial ordering problem of the travelling salesman problem (TSP) and provide extensive testing on benchmark datasets [25].