Mixed torus and hypercube multi-rank tensor expansion method
AssigneeShanghai Redneurons Co., Ltd
InventorsYUEFAN DENG, Peng Zhang
Life of this patent
Estimating the term…
Abstract
The present invention provides a mixed torus and hypercube multi-rank tensor expansion method which can be applied to the communication subsystem of a parallel processing system. The said expansion method is based on the conventional torus and hypercube topologies. A mixed torus and hypercube multi-rank tensor expansion interconnection network is built up by means of supernodes equipped with expansion interfaces. This method not only provides more bisection bandwidth to the entire system but also improves the long-range communication and global operations. Affirmatively, this expansion method can achieve better scalability and flexibility for the parallel system for a given system size.
Claims14
A method of a mixed torus and hypercube multi-rank tensor expansion
- provided to integrate multiple supernodes with expansion interfaces into a larger-scale parallel processing system, wherein,
- the supernode comprises a plurality of processing nodes and Axon nodes,
- the internal processing nodes in the same supernode are interconnected as a multi-dimensional torus to form an internal interconnection network, that is responsible for the internal communication between internal processing nodes;
- the Axon nodes connecting a subgroup or the whole of the processing nodes in the same supernode are interconnected as a multi-dimensional torus to form an external interconnection network, handling the external communication between supernodes.
A method of a mixed torus and hypercube multi-rank tensor expansion provided to integrate multiple supernodes with expansion interfaces into a large-scale parallel processing system, wherein,
- all of the internal processing nodes in all of the supernodes are interconnected as a multidimensional torus to form a trunk interconnection network, responsible for the communications between processing nodes,
- one or more Axon nodes are added to each supernode which connects a subgroup of internal processing nodes in the said supernode and provides external interfaces
- all of the external interfaces are interconnected as a multi-dimensional torus to form an external interconnection network, handling the external communication between supernodes.
Dependencies are inferred from the claim text.