After using SSAS cubes on a project recently, I’m not sure I would do it again. Some of my observations:
The performance benefits in SSAS come from compressed, columnar storage with batch computation; this reduces I/O and limits the need for maintaining various pre-built aggregate tables and covering indexes for reporting queries. Column stores are now available in mainstream relational databases: Oracle and SQL Server have column stores, and there is an open-source column store extension for PostgreSQL (citus_fdw). There’s also MonetDB, and Greenplum was recently open-sourced too.
SQL and database design knowledge are more ubiquitous than MDX and cube design. For anything non-trivial, the learning curve of cube design and MDX is going to eclipse the benefits. It doesn’t help that MDX sort of looks like SQL but is fundamentally different.
Analytics tools (Tableau, Qlik, MicroStrategy, etc.) support cubes but tend to treat them as second-class. Also these tools often provide their own in-memory OLAP engines with columnar storage, so what a cube gives you is duplicative.
Two other sources worth mentioning:
“Cubes are 1980’s technology” “As a result of columnstore performance, the customer retired their SSAS infrastructure.”The exception case might be, if you’re planning to commit to an all-Microsoft stack with Excel pivot tables or SSRS as your front-end, cubes might still make sense. But for custom webapps or a standalone analytics tool, cubes are probably not the right answer anymore.