That I cannot do, there is no single ideal size/scenario.

However: the latest databricks version is a good choice (10.0 or latest LTS for production jobs).

For data jobs, the write optimized nodes are a good choice as they can use delta cache.

For online querying: databricks sql.

I myself use the cheapest node type which handles the job, and that depends on which spark program I run. So I use multiple cluster configurations.

I even run upsert jobs with a single worker on a table of over 300 million records, works fine depending on the amount of data which has to be rewritten.

It depends on filters, transformations etc on these 300 million records.