[HIVE-20983] Vectorization: Scale up small hashtables, when collisions are detected

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 4.0.0-alpha-1
• Labels: hive-3.2.0-candidate pull-request-available
• Assignee: Mustafa İman
• Reporter: Gopal Vijayaraghavan
• Created: 29/Nov/18 08:02
• Updated: 01/Dec/22 09:12
• Resolved: 18/Sep/19 23:51

Description

Hive's hashtable estimates are getting better with HyperLogLog stats in place, but an accurate estimate does not always result in a low number of collisions.

The hashtables which contain a very small number of items tend to lose their O(1) lookup performance where there are collisions. Since collisions are easy to detect within the fast hashtable implementation, a rehashing to a higher size will help these small hashtables avoid collisions and go back to O(1) perf.

Issue Links

• links to: GitHub Pull Request #771