Issue type:

• 🐌 Performance issue

Description:

In order to resolve the problem where doing direct joins is cheaper than doing bind joins (#548 and #1196),
we introduced a heuristic (cb8d276) that only uses bind joins if the smallest stream is much smaller than the largest stream.

Currently, this heuristic uses the parameter 60 (2543e86), which seems to optimize most queries. However, for BSBM queries 2, 5, and 10, results showed that parameter 100 performed significantly better. But for other queries, this 100 value slowed down things again. This shows that this simplistic cost is missing some parameters. (WatDiv C1 requires <= 61, Ruben's query requires > 55)

Some time later, after adding isRemoteAccess (db38811), I managed to make BSBM 2 and 5 faster. But that made WatDiv S3 and S4 slightly slower. But given the overall positive impact, it was worth it.

The goal of this issue is to come up with a new cost model that includes these hidden parameters. This could include checking the number of join entries, the number of overlapping variables, the join selectivity, ...

Environment: