This message was deleted.

Ingestion rate per task seems largely based on the number of dimensions and their cardinality. Have a look at this table from some benchmarking which was done here