[KYLIN-3656] Improve HLLCounter performance

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: all
• Fix Version/s: v2.6.0
• Assignee: Chang chen
• Reporter: Chang chen
• Created: 31/Oct/18 12:59
• Updated: 12/Jan/19 06:41
• Resolved: 15/Nov/18 12:22

Description

The current HLLCounter implementation has some room to improve performance, as we find in our product environment.

1.  Create HLLCounter from another HLLCounter, we can copy register instead of merge
2. To compute harmonic mean in the HLLCSnapshot, we could
   1. using table to cache all 1/2^r  without computing on the fly
   2. remove floating addition by using integer addition in the bigger loop
   3. remove branch, e.g. needn't checking whether registers[i] is zero or not, although this is minor improvement.

Issue Links

• links to: GitHub Pull Request #345