distributed DBMSs. When a relation needs a lot of processing power and 
  a higher disk access bandwidth , the relation can be declustered as a hor-
  izontally partitioned relation and located in element DBMSs. When a 
  relation is frequently accessed, some replicas of the relation can be made 
  and located in element DBMSs. However, in the current implementation, 
  the replicated relation can be used only for the global map, that is, for 
  server DBMSs. 

  Relations can be located in main memory and/or secondary memory in 
  an element DBMS. Relations which are located only in main memory are 
  temporary relations. Quasi main memory relations both in secondary 
  memory and in main memory provide guarantees that the modifications 
  are reflected in the secondary memory. 

Query Processing 

  There are two types of command for query processing: primitive com-
  mands and KQL, a query language based on extended relational algebra. 
  Primitive commands are the lowest operations for relations, and can treat 
  relations efficiently. KQL is syntactically like KL1. New operations can 
  be defined temporarily in a query. 

  A query in KQL is translated into sub-queries in intermediate operations 
  for extended relational algebra, and is submitted to the relevant element 
  DBMSs. A query in primitive commands is submitted to the relevant 
  element DBMSs. The query is processed as a distributed transaction 
  among the relevant element DBMSs, and is finished under the control of 
  a two-phase commitment protocol. 

Parallel Processing 

  Kappa-P parallel processing takes the form of inter-cluster parallelism 
  among element DBMSs and intra-cluster parallelism in an element DBMS. 
  Inter-cluster parallelism provides more processing power, but also in-
  creases communication overhead. The trade-off is between processing 
  power and communication overhead. 

  Intra-cluster parallelism is suitable for a DBMS manipulating large amounts 
  of data. Kappa-P uses the parallelism for internal processing of an ele-
  ment DBMS, for instance parallel processing by tuple stream, operations 
  for set, and index operations of temporary relations. 


					- 77 -