The mediator-wrapper approach to integrate data from heterogeneousdata sources has usually been centralized in the sense that a singlemediator system is placed between a number of data sources and theapplications. As the number of data sources increases, the centralizedmediator architecture becomes a bottleneck. This paper presents anarchitecture for composable and distributed mediator servers, defined interms of other mediator servers. The modularity of composable mediatorsallows to build larger systems of distributed mediators integrating manydata sources, without the need to maintain a global schema. Composablemediators furthermore provide data independence by allowing locality ofchanges in both submediators and data sources. However a problem with adistributed and composable mediator architecture is that the queryperformance may degrade as the number of mediators increases. Wedescribe some challenges for processing queries in this type ofenvironment, and propose a distributed query decomposition algorithmthat eliminates some of the overhead of logical mediator composition.For certain mediator compositions it produces distributed query planswhose inter-mediator data flow is optimal with respect to the query butis different from the logical interdependencies between the involvedmediators. Experimental results show that this strategy improves thequery performance and allows an increase of the number of mediatorswithout query performance degradation
展开▼