[
https://issues.apache.org/jira/browse/LUCENE-8868?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16867327#comment-16867327
]
Ignacio Vera commented on LUCENE-8868:
--------------------------------------
I have some doubts with this optimisation in 1D. There is no much difference in
storage cost as runLen will do a good job compressing the values and we are
adding some overhead at indexing time as we need to compute cardinality. All
and all it does not hurt too much but just want o raise that concern.
> New storing strategy for BKD tree leaves with low cardinality
> -------------------------------------------------------------
>
> Key: LUCENE-8868
> URL: https://issues.apache.org/jira/browse/LUCENE-8868
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Ignacio Vera
> Priority: Major
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Currently if a leaf on the BKD tree contains only few values, then the leaf
> is treated the same way as it all values are different. It many cases it can
> be much more efficient to store the distinct values with the cardinality.
> The strategy is the following:
> 1. When writing a leaf block the cardinality is computed.
> 2. Perform some naive calculation to compute if it is better to store the
> leaf as a low cardinality leaf. The storage cost are calculated as follows:
> * low cardinality: leafCardinality * (packedBytesLength - prefixLenSum + 2)
> where two is the estimated size of storing the cardinality. This is an
> overestimation as in some cases you will only need one byte to store the
> cardinality.
> * High cardinality: count * (packedBytesLength - prefixLenSum). We are not
> taking into account the runlen compression.
> 3. If the tree has low cardinality then we set the compressed dim to -2. Note
> that -1 is when all values are equal.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
