How to tackle the "Curse of Dimensionality"?

[xiaobao520123]

When Mondrian connects to a JDBC, it becomes an ROLAP, meaning it computes MDX resultset on the fly.
This behavior leads to a serious problem, which is called "Curse of Dimensionality".

We have many dimensions in our schema design. When adding >2 dimensions, our BI frontend, Excel, stucks in the process and essentially becomes unusable.

One solution to this, according to Mondrian's docs, is using Aggregation Table. Yet designing and building Agg Tables is hard.

Does anyone have any experience you would like to share when facing dimension explosion?

Recently I've been trying using Apache Kylin as a middleware which can help me build and manage Agg Tables. It actually uses Mondrian as a MDX query engine.

[julianhyde]

The curse of dimensionality afflicts you if you try to materialize every combination of dimensions. The materialized aggregates become larger than the raw data.

The solution is to not materialize. Use a parallel DB with good I/O throughput (aided by compression) so that you can read all of the data within the target response time.

Add caching on top of that, if you wish. And build a few summary tables for combinations of dimensions that you know are commonly used.