The integration with existing BI tools is a big deal. This eliminates the need to export a subset of data into a relational DB for OLAP.

Incremental refresh of cubes is super important especially when working with large data sets.

The ACL support and LDAP integration looks nice. Proper access control on top of BI data is an absolute requirement.

I don't see anywhere this is listed explicitly, but I would guess this works with data in HBASE not individual files stored in HDFS.

The Cube File is stored on HBase now, it could be extend to otehr KV value system.

HDFS is used to store intermediate files for building cubes.

HBase is used to store data cube and execute the query. HBase co-processor is also used for query processing.

I can't seem to find the documentation.

the document will coming soon, there are many stuff we have moving from internal wiki to kyliy.io now.

what could this mean "extremely large petabytes of data analysis." ?

A good sign of 'enterprise' software: the marketing copy was written by someone from Marketing.

And then maybe translated to english by someone from accounting

And unthinkingly approved by someone from sales.

Sounds like a botched edit, replacing "extremely large" with "petabytes of" or vice-versa.