Interconnection network and crossbar switch for the same
AssigneeHitachi, Ltd.
InventorsAkira Muramatsu, Ikuo Yoshihara, Kazuo Nakao, Takehisa Hayashi, Teruo Tanaka, Shigeo Nagashima
Life of this patent
Estimating the term…
Loading patent…
AssigneeHitachi, Ltd.
InventorsAkira Muramatsu, Ikuo Yoshihara, Kazuo Nakao, Takehisa Hayashi, Teruo Tanaka, Shigeo Nagashima
Estimating the term…
In a parallel computer including L=n.sub.1 .times.n.sub.2 .times. - - - .times.n.sub.N processor element or external devices (hereafter represented by processor elements), an interconnection network of processor elements using L.times.(1/n.sub.1 +1/n.sub.2 + - - - +1/n.sub.N) crossbar switches in total comprises N dimensional lattice coordinates (i.sub.1, i.sub.2, - - - , i.sub.N), 0.ltoreq.i.sub.1 .ltoreq.n.sub.1 -1, 0.ltoreq.i.sub.2 .ltoreq.n.sub.2 -1; - - - , 0.ltoreq.i.sub.N ; and .ltoreq.n.sub.N -1 given to each processor element as the processor element number, crossbar switches for decoding a dimensional field in a processor element number having specific position and length depending upon the number of lattice points of a particular dimension and for performing the switching operation with regard to the dimension, interconnection of n.sub.k processor elements having processor element numbers, which are different only in the k-th dimensional coordinate for arbitrary k, i.e., having processor element numbers ##EQU1## by using one of the crossbar switches, each of the crossbar switches having n.sub.k inputs and n.sub.k outputs, and the interconnection performed with respect to all (L/n.sub.k sets) of coordinates (i.sub.1, i.sub.2, - - - , i.sub.k-1, i.sub.k+1, - - - , i.sub.N) of N-1 dimensional subspace excluding the k-th dimension, the interconnection being further performed for all values of k (1.ltoreq.k.ltoreq.N).
A parallel computer constructing an N-dimensional processor lattice including L=n1.times.n2.times. . . . .times.nN processor elements and an interconnection network for the processor elements including L .times.(1/n1+1/n2+ . . . +1/nN) transforming crossbar switches and L relaying crossbar switches each associated with a one of the L processor elements, wherein an information packet can be switched between the L processor elements, each of said relaying crossbar switches comprising:
A parallel computer comprising an N dimensional processor lattice including L=n1.times.n2.times. . . . .times.nN processor elements and an interconnection network, wherein nk (k=1, 2, . . . N) is a number of a lattice point on a k-th dimensional coordinate, and the interconnection network includes Lx(i/n1+1/n2+ . . . +1/nN) coordinate transforming crossbar switches and L relaying crossbar switches corresponding to each of the processor elements, wherein:
Dependencies are inferred from the claim text.