The DIOM Query Mediation Toolkit

The main task of the DIOM distributed query manager is to coordinate the communication and the distribution of the processing of information consumers' query requests among the mediator and its associated wrappers. The general procedure for multidatabase query processing is outlined in Figure 4.

: Query steps in a distributed environment with wrappers

Query Routing
This is done by mapping the domain model terminology to the source model terminology, by eliminating null queries, which return empty results, and by transforming ambiguous queries into semantic-clean queries.

Query decomposition
The goal of query decomposition is to break down an IQL query into a collection of subqueries, each targeted at a single source. The query decomposition module takes as input the IQL query expression reformulated by the query routing module after the dynamic query source selection process, and performs the query decomposition in two phases: target split and useful subquery dependency recording. Both straightforward query decomposition and complex query decomposition are considered.

Parallel query plan generation
The goal of generating a parallel access plan for the groups of subqueries is to obtain the maximum parallelism and the best quality of cooperation in searching for query answers from multiple information sources. A parallel query plan is constructed for the modified query resulted from query decomposition. The plan is composed of single-source queries (each posed against exactly one local export schema), move operations that ship results of the single-source queries between sites, and the post processing queries that assemble the results of the single-source queries in terms of the information consumer's query request expressions.

Subquery translation and execution
The translation process basically converts each subquery expressed in terms of an IDL source model into the corresponding information producer's query language expression, and adds the necessary join conditions required by the information source system.

Query result assembly
The result assembly process involves local subquery result packaging which wraps each subquery result in terms of the DIOM internal object representation, and the mediator-level result assembly which resolves the semantic variations among the subquery results. Semantic annotation and attachment is one of the main techniques that we use for resolving the semantics heterogeneity implied in the query results.

The Diorama implementation offers two sorts of query interfaces: a simple keyword-based interface for native users and an expert query interface for sophisticated customers.

• Simple Keyword-based Queries
• Expert Query
• Expert Query Tracing