The Tukwila Data Integration System

The Tukwila data integration system introduces a number of new techniques for query reformulation, optimization, and execution.

Modules

• Client/user interface
  • This module allows the user to select a mediated schema and to enter a query.
  • The query is fed to the reformulator, which translates it into a logical plan and sends it to the later stages.
  • Once query execution is complete, the execution engine notifies the user interface, and sends back the query results.
• Reformulator
  • The reformulator takes a query over the mediated schema and translates it over available web sources, generating a logical plan for the optimizer.
  • It makes use of the data source catalog and various metadata information.
  • The for Tukwila is being developed by Rachel Pottinger.
• Data source catalog
  • The source catalog contains mappings from the available data sources to the mediated schema.
  • It also contains information about data coverage, estimated access costs, etc.
• Optimizer which supports partial and incremental optimization
  • The Tukwila optimizer has the ability to specify partial plans in the presence of uncertainty.
  • It can incrementally re-optimize a query based on new information.
  • The optimizer also generates condition-action rules for the execution system, telling it how to handle various exceptions.
  • The Tukwila optimizer is under development by Marc Friedman.
• Adaptive query execution engine
  • Tukwila's execution engine supports a number of new features:
    • Rules for handling exception conditions.
    • Incremental execution, with statistics gathering and reporting.
    • Enhanced join operators to support a more data-driven model.
    • A collector operator which supports dynamic processing of information from overlapping data sources.
  • The execution system is under development by Zack Ives.
• Data source wrappers
  • Currently wrappers are available for ODBC-compliant data sources.
  • New wrappers are under development by Hakim Weatherspoon.

Papers

• Zachary Ives, Daniela Florescu, Marc Friedman, Alon Levy, Daniel S. Weld.  An Adaptive Query Execution System for Data Integration.  To appear in SIGMOD 99.