Question

In a general one-to-many (parent-to-child) relationship, is there a significant efficiency difference between (a) putting parent_id in the child table and (b) using a pivot table of only parent_id, child_id?

NOTE: Assume Oracle if necessary, otherwise answer using a general RDBMS.

Answer 1

If by PIVOT table you mean a many-to-many link table, then no, it will only hamper performance.

You should keep parent_id in the child table.

The many-to-many link table takes an extra JOIN and therefore is less efficient.

Compare the following queries:

SELECT  *
FROM    child_table c
JOIN    child_to_parent cp
ON      cp.child = c.id
JOIN    parent p
ON      p.id = cp.parent
WHERE   c.property = 'some_property'

and this one:

SELECT  *
FROM    child_table c
JOIN    parent p
ON      p.id = c.parent
WHERE   c.property = 'some_property'

The latter one is one JOIN shorter and more efficient.

The only possible exception to that rule is that you run these queries often:

SELECT  *
FROM    child_table c
JOIN    parent_table p
ON      p.id = c.parent
WHERE   c.id IN (id1, id2, ...)

, i. e. you know the id‘s of the child rows beforehand.

This may be useful if you use natural keys for your child_table.

In this case yes, the child_to_parent link table will be more efficient, since you can just replace it with the following query:

SELECT  *
FROM    child_to_parent cp
JOIN    parent_table p
ON      p.id = cp.parent
WHERE   cp.child IN (id1, id2, ...)

and child_to_parent will be always less or equal in size to child_table, and hence more efficient.

However, in Oracle you can achieve just the same result creating a composite index on child_table (id, parent_id).

Since Oracle does not index NULL‘s, this index will be just like your child_to_parent table, but without the table itself and implied maintenance overhead.

In other systems (which index NULL‘s), the index may be less efficient than a dedicated table, especially if you have lots of NULL parents.