Parallel control technique and performance of an MPPM knowledgebase machine

摘要

A description is given of parallel control techniques andnperformance evaluations by simulation for the knowledge-base machinen(KBM) using the multiport page-memory (MPPM) and the unification enginen(UE). Relational knowledge base retrieval requires repeated applicationnof the unification-join (U-join) operations on term relations. It alsonrequires each U-join to be dynamically scheduled, observing the resultnof the previously executed U-join. Control techniques to execute thenquery process of the relational knowledge base effectively in the KBMnusing the MPPM and the UE have been investigated. For parallel executionnof coarse-grain operations, each operation is decomposed to concurrentlynexecutable fine-grain operations by partitioning input data sets. Annumber of features that depend on processor-allocation strategies havenbeen identified. In particular, it was found that query processingnresulting in repetitions of the retrieval operation, by a certainncontrol strategy, increases the amount of the total data transfer to thenhighly parallel UEs and lowers processing efficiency