If one is building two systems, each with a separate schema but with a few cross links between them, at one point does it make sense to use separate databases vs. putting all into the same database? That's a pretty broad question I realize, but just searching for any insight on the issue. Assume the systems in question are large in terms of volume and load, each more OLTP than analytic but some of the latter.
In fact I also wonder if it would ever make sense to put both systems in the same database but use DPF to isolate the load down between each system to the storage level. There is a service integration layer in front, and it could be relied on to always know to which of the two database partitions to connect to depending on the targeted table. This would appear to be very similar to just having two completely separate systems, except that when needed cross schema SQL operations can be freely used and rely on DB2's smarts about optimizing the request across the partitions. It would also seem to obviate the need to do two-phase commits at the service layer for updates that affected both schemas, assuming DB2 DPF does that internally just as well (?).
Any insight on this would be much appreciated.