Objective of Kappa-P 
The objective of Kappa-P is to provide database management facilities for 
many KIPSs, for instance natural language processing systems with electronic 
dictionaries, proof checking systems with mathematical knowledge, and ge-
netic information processing systems with molecular biological data. Kappa-
P has been developed to manage large amount of complex structured data 
efficiently. 

Nested Relational Model 
  In order to treat complex structured data efficiently, the conventional re-
  lational model must be extended. In Kappa-P, a nested relational model 
  with a set constructor and hierarchical attributes can represent complex 
  data naturally, and can avoid the unnecessary division of relations. More-
  over, the semantics of the model matches the knowledge representation 
  language Quixote, which is the upper layer of the KBMS of the FGCS 
  project. Kappa-P has charge of the database engine of this system.
 
  Term is added as a data type in order to store various types of knowledge. 
  The character code of the PIM machine is based on 2-byte code, but the 
  code wastes secondary memory space. In order to store a huge amount of 
  data, data compression and index facilities have been improved. 

Configuration 
  The configuration of Kappa-P corresponds to the architecture of the PIM 
  machine, and distinguishes inter-cluster parallelism from intra-cluster par-
  allelism. Kappa-P consists of a collection of element DBMSs located in 
  clusters. These element DBMSs cooperate in processing a query. 

  The global map of relations is managed by element DBMSs Called server 
  DBMSs. Server DBMSs manage not only the global map but also ordi-
  nary relations. Element DBMSs, except server DBMSs, are called local 
  DBMSs. Interface processes are created to mediate between application 
  programs and Kappa-P, and to receive and send messages such as queries 
  and answers. 

Data Placement 
  The placement of relations also corresponds to parallelism: inter-element 
  DBMS placement and intra-element DBMS placement.
 
  In order to use inter-cluster parallelism, relations can be located in sev-
  eral element DBMSs. A simple case is the distribution of relations like 


					- 76 -