change.log

08/11/2014 
  Added support for bitmap join indexes. Added SSD optimizations for storage engine.

02/26/2014
  Added a simple built-in data dictionary.	

02/20/2014
  Added ability to use Alenka as a lib call.

02/07/2014
  Added support for sparse datasets : one can load the data like this now : 
	A := LOAD 'lineitem.tbl' USING ('|') AS (orderkey{1}:int REFERENCES orders(1) ...
  In order to use this you need to load orders first, then lineitems. Performance of many join operations 
  is improved by 2X.


09/14/2013
  Join tables that do not fit into limited GPU memory are partitioned :
  STORE A INTO 'lineitem' BINARY SORT SEGMENTS BY orderkey PARTITION BY 2; 
  This way it becomes possible to join huge tables that normally won't fit into GPU memory.

09/05/2013 
  Filters are done in-place, host memory usage is improved.

07/04/2013
  Major join performance improvements due to addition of local segments sort option.

05/27/2013
  Added a sorted merge joins as default join type.
  Added support for left, right and outer joins.
  Tested a few TPC-H queries on a 300GB dataset.


05/10/2013
   Added support for star query joins (important for data warehousing applications)
     -look at star_schema_queries.zip
   Support for multitable joins.
   Tested on a 1TB database (Star Schema Benchmark).
   Fixed a problem with grouping of datasets that do not fit GPU.


03/22/2013
   Joins are now GPU hash joins.

02/16/2013
   Performance and bug fixes

02/10/2013
   Modified groupby op - now it is hash-based and scalable.
   Switched to Sean Baxter's binary search implementation (it is significantly faster)
   Added support for COUNT(DISTINCT field))


01/28/2013
    Bug fixes.


01/19/2013.
    Major changes to the way the strings are stored and processed by the engine.
    Implemented join on strings using Murmurhash library.
    Implemented external sort.


11/30/2012.
    Modified the database file structure for faster access to the database segments.



11/16/2012.
    I have verified that the program runs correctly for 64bit values.
    Because CUDPP Hash table supports only 32 bit keys I had to modify the hash join algorithm.