• Only SQL-type KBs can be member of a cluster.
• Documentation (definition) of the cluster is stored in the  cluster-table in the p3publications SQL database.
• The cluster consists of an originating KB (set up by the KB), and one or more automatically generated clones of this KB.
• All members have a member number. The originating member has number 0, while other members have a unique number greater than zero (normally 1,2,3 etc).
• The generated KBs have the same name as the originating KB, but are located in different SQL databases (whose names are xxx_N where xxx is the SQL database of the originating KB and N is the member number.
• All members read from one knowledge stream.
• All members use one open share in incremental mode (this is knowledge-wise the device by which the members share and exchange knowledge).

Clusters can be created with the Create Cluster-window, and they are processed via the Process Cluster-command in the Sentences-menu.

PROCESSING OF A CLUSTER

The KE creates the originating KB, the publishing network and its open share, the knowledge stream and finally defines the cluster. The cluster can then be processed (blue symbols in figure)

The originating KB reads all facts from the knowledge stream, and all sentence rules, functions and inexact rules which do not violate the requirement to keep the knowledge dependency graph acyclic. The rules, functions and inexact rules which remain unread are subsequently read by the cloned KB generated by the originating KB (with the same requirement to keep its dependency graph acyclic). (green symbols in figure).

The member KBs are opened in a round-robin way and performs the triple action of open share sentence subscription, sentence derivation and open share sentence publication, until no more sentences can be derived.

The Einstein Puzzle model and its step-by-step guide is a good intro to knowledge base clustering!.