In a multi-cell situation you have two topology options for configuring multi-cell routing preference rules for applications that exist in two cells, which we consider to be recommended best practices. Neither of these topologies employ core-group bridges and for Versions 7.x and higher it is discouraged to use core-group bridges to link cells.
The Star Topology works well if you have multiple cells and the center cell contains all on demand routers (ODR's). You would remove all ODRs from the point cells and move them into the center cell. For the sake of high availability it's a good idea to cluster the ODRs, and to keep them on separate hardware and separate power supplies, etc. You would then use the linkcells script to link the cells. This is easy, it's quick, and it's a best practice. It's highly-scalable, as you can add more cells as necessary. Discovery is also still there -- if you add applications to one cell or another, the ODR will learn of them automatically, just as if everything were in one cell.
For information on configuring a star topology please see Configuring multi-cell performance management: Star Topology
The Peer-cell Topology works well if you have two or more disjoint data centers, one cell per data center, and you want failover capability between them. In this topology, the ODRs remain in the cells; however, the two cells are not joined via core-group bridges. In front of your two cells, you have one or more load-balancers, plugins, or sprayers which are able to both preserve session affinity (as applicable), as well as equitably distribute traffic.
For more information on configuring a peer-cell topology please see Configuring multi-cell performance management: Peer-Cell Topology