[ 
https://issues.apache.org/jira/browse/KYLIN-4985?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17370190#comment-17370190
 ] 

Zhong Yanghong edited comment on KYLIN-4985 at 6/27/21, 10:37 AM:
------------------------------------------------------------------

Hi [~tianhui5], the cuboids recommended by cube planner algorithms is 
redundant. The redundancy is controlled by 
*kylin.cube.cubeplanner.expansion-threshold*.

One more thing for "get many cuboids in recommand result that never hitted by 
my history queries". If cuboid A is the parent of cuboid B, and their row 
account are similar, even when your history queries always hit cuboid B, Kylin 
should prefer to choose cuboid A to be built.

For the weighting change, could you explain more about the mathematical theory? 
At first glance, it's not follow monotonicity of the probability.


was (Author: yaho):
Hi [~tianhui5], the cuboids recommended by cube planner algorithms is 
redundant. The redundancy is controlled by 
*kylin.cube.cubeplanner.expansion-threshold*.

One more thing for "get many cuboids in recommand result that never hitted by 
my history queries". If cuboid A is the parent of cuboid B, and their row 
account are similar, even when your history queries always hit cuboid B, Kylin 
should prefer to choose cuboid A to be built.

For the weighting change, could you explain more about the mathematical theory? 
At first glance, it's not follow monotonicity.

> optimize kylin planner by delete unnecessary cuboids
> ----------------------------------------------------
>
>                 Key: KYLIN-4985
>                 URL: https://issues.apache.org/jira/browse/KYLIN-4985
>             Project: Kylin
>          Issue Type: New Feature
>            Reporter: tianhui
>            Priority: Major
>
> When I use Kylin Planner, I can get many cuboids in recommand result that 
> never hitted by my history queries. I think it maybe unnecessary, so I delete 
> the unhitted cuboids.
> In addition, I change row count by weighting of 1/sqrt(hit probability) 
> before execute plan algorithm.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to