US 8,949,577 B2Granted

Performing a deterministic reduction operation in a parallel computer

AssigneeInternational Business Machines Corporation

InventorsCharles J. Archer, Michael A. Blocksome, Joseph D. Ratterman, Brian E. Smith

Life of this patent

Estimating the term…

Loading patent…

Abstract

A parallel computer that includes compute nodes having computer processors and a CAU (Collectives Acceleration Unit) that couples processors to one another for data communications. In embodiments of the present invention, deterministic reduction operation include: organizing processors of the parallel computer and a CAU into a branched tree topology, where the CAU is a root of the branched tree topology and the processors are children of the root CAU; establishing a receive buffer that includes receive elements associated with processors and configured to store the associated processor's contribution data; receiving, in any order from the processors, each processor's contribution data; tracking receipt of each processor's contribution data; and reducing, the contribution data in a predefined order, only after receipt of contribution data from all processors in the branched tree topology.

Claims12

Dependent claims

An apparatus for performing a deterministic reduction operation in a parallel computer, the parallel computer comprising a plurality of compute nodes, each compute node comprising a plurality of computer processors and a CAU (Collectives Acceleration Unit), the CAU coupling computer processors of compute nodes to one another for data communications in a cluster data communications network, the apparatus comprising a computer processor, a computer memory operatively coupled to the computer processor, the computer memory having disposed within it computer program instructions that, when executed by the computer processor, cause the apparatus to carry out the steps of:

organizing three or more processors of the parallel computer and a CAU into a highly branched tree topology, wherein the CAU is a root of the highly branched tree topology and each processor in the highly branched tree topology is a direct child of the root CAU;
establishing a receive buffer comprising a plurality of receive elements, each receive element associated with one of the processors in the highly branched tree topology and configured to store the associated processor's contribution data;
receiving, by the root CAU in any order from the processors in the highly branched tree topology, each processor's contribution data;
tracking receipt of each processor's contribution data wherein tracking receipt of each processor's contribution data further comprises maintaining a count of the number of processors from which contribution data has been received;
and
reducing, by the root CAU, only after receipt of contribution data from all processors in the highly branched tree topology, the contribution data stored in the receive buffer in a predefined order wherein reducing the contribution data stored in the receive buffer in a predefined order further comprises reducing the contribution data only after the count is equal to the number of processors in the highly branched tree topology.

3 more elements

A computer program product for performing a deterministic reduction operation in a parallel computer, the parallel computer comprising a plurality of compute nodes, each compute node comprising a plurality of computer processors and a CAU (Collectives Acceleration Unit), the CAU coupling computer processors of compute nodes to one another for data communications in a cluster data communications network, the computer program product disposed upon a computer readable storage medium, wherein the computer readable storage medium is not a signal, the computer program product comprising computer program instructions that, when executed, cause a computer to carry out the steps of:

organizing three or more processors of the parallel computer and a CAU into a highly branched tree topology, wherein the CAU is a root of the highly branched tree topology and each processor in the highly branched tree topology is a direct child of the root CAU;
establishing a receive buffer comprising a plurality of receive elements, each receive element associated with one of the processors in the highly branched tree topology and configured to store the associated processor's contribution data;
receiving, by the root CAU in any order from the processors in the highly branched tree topology, each processor's contribution data;
tracking receipt of each processor's contribution data wherein tracking receipt of each processor's contribution data further comprises maintaining a count of the number of processors from which contribution data has been received;
and
reducing, by the root CAU, only after receipt of contribution data from all processors in the highly branched tree topology, the contribution data stored in the receive buffer in a predefined order wherein reducing the contribution data stored in the receive buffer in a predefined order further comprises reducing the contribution data only after the count is equal to the number of processors in the highly branched tree topology.

3 more elements

Dependencies are inferred from the claim text.