Data mediation aims to access or evaluate queries on heterogeneous data sources. It is a subdomain of data integration with a focus on performing efficient data retrieval.
The mediators are the mediation systems.
Query execution plans
They usually take as input a query execution plan and returns its answers. This input plan is defined by a logical plan, which is a tree where the nodes are operators on the data. For example in relational plan, an operator can be a projection, a join or a selection. Performing this plan requires to choose an implementation for each operator of the logical plan. We obtain a physical plan, when the implementation of every operator of the plan are fixed.
Cost of query plan
In order to optimize the execution of a query plan P, the mediator have to find the cheapest physical plan among the space of every physical plan that are equivalent to P. It requires a cost model on the physical plans. Usually, the cost of a logical plan is defined as the minimal cost of its implementations.
The monotony assumption on cost model is the fact that if a plan P' is obtained from the plan P by replacing a subplan P1 of P by a cheaper and equivalent subplan P2, then P' is cheaper than P DBLP:conf/sigmod/FlorescuLMS99.