We know from performance testing that index intersection plans are faster than single-solution plans when
- The size of the intersection is small.
- The amount of data that would need to be fetched by the single-index solutions is large, so that
- this data would exceed the available memory, or is generally not in the working set.
However, our testing also shows that index intersection plans may not be chosen by the ranker in these cases. We should consider tweaking plan ranking such that the ranker is more likely to choose intersection plans.
- is duplicated by
-
SERVER-15125 Index intersection amount of work should be measured incorporating all its phases
- Closed
-
SERVER-17821 Query planner prefers suboptimal index to using index intersection
- Closed
-
SERVER-28587 MongoDB aggregation query slowness
- Closed
- is related to
-
SERVER-20619 Statistics-based query optimization
- Backlog