icomefromthenet / mysqlfastintervallookup

RITree implementation in mysql for fast interset interval lookups.

Examples of Allens Relations

Using the timslot_slots table for any two intervals

Let p1 = opening_slot_id, p2 = closing_slot_id Let q1 = opening_slot_id, q2 = closing_slot_id

Allens Relations

1. equals -> allens/equals.mysql
2. before -> allens/before.mysql
3. after -> allens/before.mysql
4. meets -> allens/meets.mysql
5. meets-by -> allens/meets.mysql
6. overlaps -> allens/overlaps.mysql
7. overlaps-by allens/overlaps.mysql
8. during -> allens/during.mysql
9. includes -> allens/during.mysql
10. starts -> allens/starts.mysql
11. stared-by -> allens/starts.mysql
12. finishes -> allens/finishes.mysql
13. finished-by -> allens/finishes.mysql

Combinaions

1. Aligns -> combine/aligns.mysql
2. Excludes -> combine/excludes.mysql
3. Fills -> combine/fills.mysql
4. Intersects -> combine/intersects.mysql
5. occupies -> combine/occupies.mysql

The closed:open interval format

There are four ways to represent an interval depending on the inclusion of the opening and closing values.

The generally prefered format in database is closed:open a closing interval is always closing value +1 which actually the opening of the next range.

Two intervals with 5 mintues each open:open (0:6) closed:closed [1:5] open:closed (0:5] close:open [1:6)

The closed:open is prefered as it allows a continous range of intervals to be easily searched. If closed:closed is used the above range would be [1:5] and the next range would be [6:10]. Its easier to build a continous set if the intervals p[1:6) & q[6:11) we can match 'p2=q1'. If we used closed-closed we would need to know the length of the clock tick between intervals so we could write a query like '(p2 - q1) = clock-tick'

RI Tree

The Relation Interval Tree is an adaption of and Interval Tree structure into a relational context. The algorithm was outlined out in a paper called Managing Intervals Efficiently in Object-Relational Databases.

I have implement their fixed height tree to speed up my intersection queries.

The table timeslot_slots_tree tree has been modifed to support the RITree. It is based on the original timeslot_slots table.

An example of every day query can be found in tree/basic.mysql. A comparision query of common interval search can be found in tree/normal.mysql.

Using a low teir VPS hosted with Linode, running both queries to find intersection of timeslots within a particular calendar day.

Duration(sec): Normal : 0.07690900 RITree : 0.00163325

The RITree was able to significantly speed up the intersects query.