hive-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Namit Jain (JIRA)" <>
Subject [jira] [Commented] (HIVE-3652) Join optimization for star schema
Date Mon, 05 Nov 2012 04:19:13 GMT


Namit Jain commented on HIVE-3652:

[~amareshwari], do you think it would be possible to get a cheap implementation with a single
mapper performing
multiple dimension joins one after the other ? I think, we should start with a few assumptions:

1. The user will not give any map-join hints. Let us derive thin info. from the statistics,
as Mark suggested.
2. This should be a single map-only job (no reducer).
3. As a first cut, it might be useful to optimize the query:

   select /*+ MAPJOIN(b,c) */ from FACT a join DIM1 b on a.k1=b.k1 JOIN DIM2 c on a.k2=c.k2

The above query should not have a back-up task in case the map-join fails, and it should be
as a single map-only job. Once that is done, using cost to convert join into map-join can
be explored independently.

> Join optimization for star schema
> ---------------------------------
>                 Key: HIVE-3652
>                 URL:
>             Project: Hive
>          Issue Type: Improvement
>          Components: Query Processor
>            Reporter: Amareshwari Sriramadasu
>            Assignee: Amareshwari Sriramadasu
> Currently, if we join one fact table with multiple dimension tables, it results in multiple
mapreduce jobs for each join with dimension table, because join would be on different keys
for each dimension. 
> Usually all the dimension tables will be small and can fit into memory and so map-side
join can used to join with fact table.
> In this issue I want to look at optimizing such query to generate single mapreduce job
sothat mapper loads dimension tables into memory and joins with fact table on different keys
as well.

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see:

View raw message