PURPOSE: To reduce the entire contention arbitration time and a transfer path for various data by forming a large scale matrix switch with sets of lots of small scale matrix switches and processing contention arbitration in parallel through divided processing.
CONSTITUTION: Groups of divided request generating sources R1-R8 and resources S1-S8 are mutually connected by each of small scale matrix switches G(1, 1)-G(4, 4) respectively. Through the constitution above, access requests and/or data from the request generating sources R1-R8 are directly transferred to the matrix switches G(1, 1)-G(4, 4) interconnected by input side connection lines L1-L8, and the access requests and/or data from the matrix switches G(1, 1)-G(4, 4) to the resources S1-S8 are directly transferred to the resources S1-S8 interconnected by output side connection lines LO1-LO8. Thus, number of passing cross points is decreased and the contention arbitration is implemented in parallel by using each small scale matrix switch, then the entire processing speed is quickened.
YAMANAKA NAOAKI
DOI YUKIHIRO
GENDA KOICHI