Investigate the performance of BigFUN queries against columnar collections

A run was conducted to test the performance of columnar collections. All queries were slower compared to row collections. That's expected when all columns are requested; however, query BF08 and BF14 project only a few columns (i.e., not SELECT *) but it was slower.

Initial look:

• The plans for both queries were correct and showed that only the required columns were projected.  
• Oceanus nodes have large memory (64GB) – even with smaller buffer cache, the file system buffer cache could play a big role here in minimizing the I/O cost
  • Also, there're two SSD's on each node. I/O may not be a bottleneck.
• The assembly cost was too high – given the number of records that needed to be assembled (probably the main culprit)