Distributed query optimization in loosely coupled multidatabase systems

Query evaluation for distributed heterogeneous relational databases

“…Many theoretical solutions have been proposed, and different implementations have provided varying approaches to these problems [8,14,16,17,31,37]. In [14], a hybrid algorithm was developed to transform a left deep join tree into a balanced bushy join tree based on a simple cost analysis, so that the overall response time can be reduced.…”

“…Moreover, in that method, only the communication costs are considered (see Appendix B of [17]), and the load measurements are not addressed at all. Similarly, in [37], a fourth method was suggested for finding an optimal join sequence with node allocation from a query graph in an exhaustive but overhead-distributed manner. But tree balance was not considered.…”

Decoupled query optimization for federated database systems

“…But in general, the execution costs may also depend on transient system issues including current loads and temporal administrative policies [45], and hence the cost estimation process must be federated in a manner reflective of the query processing, with cost estimates being provided by the sites that would be doing the work. Many of these decouplings have been studied before individually in the context of distributed, heterogeneous or federated database research [41,15,38]. However, to the best of our knowledge, complete decoupling of cost estimation, which requires the optimizer to communicate with the sites merely to find the cost of an operation, has not been studied before.…”

“…Garlic uses exhaustive dynamic programming to find the optimal plan, and though we are not aware of any explicit work in this area that uses the other optimization techniques, it should not be difficult to extend those for this scenario. [15,38,41] discuss query optimization issues in a loosely coupled federated system such as modelled by the top row. From the query optimization perspective, the main focus of [15,38] is on incorporating the knowledge of statistics and system parameters gained while executing the query at run-time, into the query plan.…”

“…They use a statistical decision mechanism that schedules the query plan a join at a time. [41] propose a mechanism for a loosely coupled multi-database system where all the underlying databases cooperate to find the optimal plan. This algorithm is not very suitable for a federated database system as the environment may not be completely cooperative and also, the number of messages that are exchanged between the underlying data sources increases exponentially with the size of the query.…”