OLAP servers support fast multidimensional analysis of warehouse data. ROLAP servers store data in relational tables and generate SQL; MOLAP servers store data in pre-computed multidimensional arrays (data cubes).
ROLAP (Relational OLAP)
Architecture (in words): Data warehouse (RDBMS with star/snowflake schema) ↔ ROLAP server (metadata + SQL generation, aggregation navigation) ↔ Front-end / client tools.
- Sits between the relational back end and client tools; translates each OLAP request (slice, dice, drill-down) into SQL queries.
- Uses star or snowflake schemas and summary (aggregate) tables.
- Advantages: handles very large data volumes; uses existing RDBMS security and tools; no separate cube storage.
- Disadvantages: slower queries because aggregations are often computed on the fly; complex SQL.
- Examples: Oracle (with OLAP options), MicroStrategy.
MOLAP (Multidimensional OLAP)
Architecture: Data warehouse / sources → (load and pre-aggregate) → MOLAP server with multidimensional database (data cube) ↔ Front-end tools.
- Data stored in array-based multidimensional storage; aggregates computed in advance.
- Advantages: very fast query response; good for complex calculations and slicing/dicing.
- Disadvantages: limited data volume; cube must be rebuilt/refreshed; storage waste for sparse data (handled using sparse-matrix compression).
- Examples: Essbase, Microsoft Analysis Services (MOLAP mode).
Comparison
| Point | ROLAP | MOLAP |
|---|---|---|
| Storage | Relational tables | Multidimensional cube |
| Speed | Slower | Faster |
| Data volume | Very large | Moderate |
| Aggregation | Mostly at query time | Pre-computed |
| Query language | SQL | Cube operations (e.g. MDX) |
HOLAP (Hybrid OLAP) combines both: detailed data in relational tables, aggregates in a MOLAP cube.