shubhamvishu commented on PR #12716:
URL: https://github.com/apache/lucene/pull/12716#issuecomment-1781810333

   @mikemccand @bruno-roustant  So I also ran the FST construction time 
benchmarks on `wikimediumall` index using `IndexToFST` over a couple of 
combinations of tweakable parameters to see the effect. Following are the 
results :
   
   | Golden ratio Bit mixing | Rehash ratio (`2/3`) | Rehash ratio (`3/4`) | 
Quadratic Probing | Linear Probing | No. of iterations | Avg Time (seconds) | 
Comments |
   
|:-----------------------:|:------------------:|:------------------:|:-----------------:|:--------------:|-------------------|:------------------:|:--------:|
   |            No           |         Yes        |          -         |        
Yes        |        -       |         10        |       89.5471      | 
**Baseline** |
   |           Yes           |         Yes        |          -         |        
Yes        |        -       |         10        |       89.6913      |          
|
   |           Yes           |         Yes        |          -         |        
 -         |       Yes      |         10        |       89.8202      |          
|
   |           Yes           |          -         |         Yes        |        
Yes        |        -       |         10        |        98.15       |          
|
   |           Yes           |          -         |         Yes        |        
 -         |       Yes      |         10        |      103.4483      |    
**PR**    |
   
    
   Then I also tried rehash ratio of `1/2` (just to explore ahead) to see its 
impact, that seem to have slightly better numbers (best amongst the tried ones).
   
   | Golden ratio Bit mixing | Rehash ratio (`1/2`) | Quadratic Probing | 
Linear Probing | No. of iterations | Avg Time (seconds) | Comments |
   
|:-----------------------:|:------------------:|:-----------------:|:--------------:|-------------------|:------------------:|:--------:|
   |           Yes           |         Yes        |        Yes        |        
-       |         10        |       89.1806      |          |
   |           Yes           |         Yes        |         -         |       
Yes      |         10        |       89.2009      |          |
   
   I'm not aware of the rationale behind choosing rehash ratio of `2/3`, but 
should we try making it `1/2`? That might have few consequences which I am 
unaware of. Keen to know your thoughts on this. Thanks!


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org

Reply via email to