Abstract: In this paper, we propose a parallel architecture for very large knowledge base (?) Major features of this architecture are:a, integrated abstract model eliminates data inconsistency and fun(?) redundancy; b. combination of plan generating and query retrieving speeds up processing, c. data flow exe(?) mechanism of multi-equi-join renders our architecture extremely high parallel; d. multi-level memory hrer(?) supports large scale knowledge management: e tight coupling approach reduces communication overhead; and f.set-oriented strategy replaces tuple-oriented strategy.