There is no reason why ClickHouse, Druid and Pinot couldn’t copy successful serialization, compression or query execution techniques from each other realtively easily. I was talking about the potential limit, but all three systems are very far from that potential limits at the moment for the majority of data schemas and query types.

Why couldn’t you add more nodes in ingestion service and ingest more data?

I think ingestion rate doesn’t depend on data format, and vice versa. Because it’s just accumulating rows (row-based) and then converting them into columnar format in background. And so surely there are no reasons why Pinot and ClickHouse couldn’t have as good ingestion as Druid, if they don’t already. Or better. Because I think Druid is also far from the optimum here, too. As well as ClickHouse and Pinot.