What is a Multi-dimensional Index Structure?
Multi-dimensional index structures, told as a story about tagging an inventory card with zone, expiry, and category all at once.
The driver wants boxes that are both in zone B and expiring this week.The sorter has to search every shelf from the start.
One index speeds up searches on only one key.A zone index doesn't speed up a search by expiry date.
A multi-dimensional index puts several tags on one card at once.Even a mix of zone, expiry, and category is found without searching every shelf.
Different search jobs call for different shapes of multi-dimensional index.Location searches use an R-tree; yes/no filters use a bitmap.
More tags means every new box takes longer to file.There's also a vector index that finds data with a similar meaning.
A multi-dimensional index structure is built to search quickly by several attributes at once, not just one.
Unlike a plain index sorted by one column, a multi-dimensional index handles several conditions together — through space (R-trees, k-d trees), presence/absence (bitmaps), or vector similarity. The trade-off: more indexes raise write cost, and too many dimensions bring on the 'curse of dimensionality', where efficiency actually drops.
When grown-ups say it
- Composite Index
- An index built from several columns together. Column order affects search performance too.
- R-tree
- An index that narrows space by nested boxes. Common for map and coordinate data.
- Bitmap Index
- An index stacking yes/no as 0s and 1s. Fast when there are few distinct values.
- Curse of Dimensionality
- Too many dimensions, and index searches slow down. The index becomes less effective, not more.
- Spatial Index
- An index searched by location. The R-tree is a typical implementation.
- Vector Index
- An index that finds things by meaning. Used for embedding search in the AI field. → a warehouse that searches by meaning
- Write Overhead
- The extra work every time a new box is filed. More indexes mean slower inserts.
- Covering Index
- When the index alone answers the question. No trip to the shelf needed — even faster.