Ongoing Work

The current version of K2 is 0.5 alpha. Here's a list of the things we're currently doing or plan to do for upcoming releases, in no particular order:

• Integrate access to password-protected data sources.

• Administrators can view information about the server, bring the server to a state where it's safe to stop, and force client connections to close. Future requirements include the ability to:

  • Bring the server completely down, cleanly if possible, and forcibly if necessary.
  • Check the progress of currently executing queries.
  • Add and remove function definitions.
  • Force cached driver results to be re-evaluated.
  • Change client connection restrictions.

• Improve the cost model portion of the optimizer. In particular:

  • K2 represents queries internally as an abstract syntax tree. Most tree nodes are currently giving accurate estimates, and the rest use default values or make imperfect but reasonable guesses. They need to be improved.
  • A basic framework is in place that allows data drivers to provide cost estimates, but this needs to be expanded and finalized, and the existing drivers need to begin providing accutare estimates.
  • The cost estimates are currently used to reorder conditional nodes, and could be used to reorder loops. A better approach might be to generate the complete set of competing query plans and estimate the cost of each. More research needs to be done in this area.
  • Drivers that export data source entry points as variables can provide information about indexes that exist in the underlying source. This information is not currently used.

• Allow driver-exported data source entry points to have more than one type.

• Investigate the possibility of blending the optimization and execution phases. For example, when it is possible to determine that a portion of the query is optimal, but the best plan for the query as a whole is uncertain, what if we execute the portion we're sure of and then reconsider the rest? The new certainty about the result of the subquery may enable us to decide among the query plans for the remainder of the query.

• Implement disk storage of partial results, so the entire result doesn't need to be held in memory.

• Implement disk storage of cached driver query results, so they don't need to be held in memory. Also, create a way to specify an "expiration time" for cached results, on a per-driver basis.

• Complete the implementation of all combinations of lazy (pipelined) and parallel (multithreaded) query evaluation.

• Complete support for OQL, including anonymous sets, group by, having, and order by. Complete support for ODL, including relationships in interfaces.

• Add XML as an output format, and develop a data driver for XML.

• Develop a CORBA data driver.

• Remove K2's dependency on JGL collections by using the collection classes provided with Java™ 1.2.

[K2 Front Page]