초록 ▼

Methods, systems, and products are disclosed for broadcasting collective operation contributions throughout a parallel computer. The parallel computer includes a plurality of compute nodes connected together through a data communications network. Each compute node has a plurality of processors for u

Methods, systems, and products are disclosed for broadcasting collective operation contributions throughout a parallel computer. The parallel computer includes a plurality of compute nodes connected together through a data communications network. Each compute node has a plurality of processors for use in collective parallel operations on the parallel computer. Broadcasting collective operation contributions throughout a parallel computer according to embodiments of the present invention includes: transmitting, by each processor on each compute node, that processor's collective operation contribution to the other processors on that compute node using intra-node communications; and transmitting on a designated network link, by each processor on each compute node according to a serial processor transmission sequence, that processor's collective operation contribution to the other processors on the other compute nodes using inter-node communications.

대표청구항 ▼

1. A method of broadcasting collective operation contributions throughout a parallel computer, the parallel computer comprising a plurality of compute nodes connected together through a data communications network, each compute node having a plurality of processors for use in collective parallel ope

1. A method of broadcasting collective operation contributions throughout a parallel computer, the parallel computer comprising a plurality of compute nodes connected together through a data communications network, each compute node having a plurality of processors for use in collective parallel operations on the parallel computer, each collective parallel operation comprising a message passing computer program instruction executed simultaneously by all processors in all compute nodes of the parallel computer, the method comprising: transmitting, by each processor on each compute node responsive to each processor's simultaneous execution of a message passing computer program instruction of a collective parallel operation, that processor's collective operation contribution to the other processors on that compute node using intra-node communications;transmitting on a designated network link, by each processor on each compute node according to a serial processor transmission sequence, that processor's collective operation contribution to the other processors on the other compute nodes using inter-node communications, wherein the serial processor transmission sequence specifies the order in which each processor on a compute node may access the designated network link for transmitting a collective operation contribution through the network to another compute node,wherein each processor's collective operation contribution comprises data contributed by that processor to effect the collective parallel operation; andwherein transmitting on a designated network link comprises:determining, by each processor on that compute node, that processor's position in the serial processor transmission sequence, andtransferring, by each processor on that compute node, that processor's collective operation contribution to the other processors on the other compute nodes in dependence upon that processor's position in the serial processor transmission sequence. 2. The method of claim 1 wherein: each compute node has a single designated network link for transmitting the collective operation contributions of each processor on that compute node. 3. The method of claim 1 wherein: each compute node has a plurality of designated network links for transmitting the collective operation contributions of each processor on that compute node, each designated network link for each compute node uniquely corresponding to one of a plurality of network link domains for that compute node in which a subset of the processors for that compute node are organized, each network link domain having a serial domain processor transmission sequence;the method further comprises assigning each processor on each compute node into one of the network link domains for that compute node, including assigning that processor a position in the serial domain processor transmission sequence for that network link domain; andtransmitting on a designated network link, by each processor on each compute node according to a serial processor transmission sequence, that processor's collective operation contribution to the other processors on the other compute nodes using inter-node communications further comprises performing in parallel for each network link domain:determining, by each processor assigned to that network link domain on that compute node, that processor's position in the serial domain processor transmission sequence, andtransferring on the designated network link corresponding to that network link domain, by each processor in that network link domain on that compute node, that processor's collective operation contribution to the other processors on the other compute nodes in dependence upon that processor's position in the serial domain processor transmission sequence. 4. The method of claim 1 wherein the intra-node communications are carried out using shared memory operations, direct memory access ('DMA') operations, or message passing operations. 5. The method of claim 1 wherein broadcasting collective operation contributions is carried out as part of a gossiping communications pattern. 6. The method of claim 1 wherein the plurality of compute nodes are connected together through a plurality of data communications networks, at least one of the data communications networks optimized for point to point data communications, at least one of the data communications networks optimized for collective operations. 7. A parallel computer for broadcasting collective operation contributions throughout a parallel computer, the parallel computer comprising a plurality of compute nodes connected together through a data communications network, each compute node having a plurality of processors for use in collective parallel operations on the parallel computer, each collective parallel operation comprising a message passing computer program instruction executed simultaneously by all processors in all compute nodes of the parallel computer, the parallel computer comprising computer memory operatively coupled to the processors of the parallel computer, the computer memory having disposed within it computer program instructions capable of: transmitting, by each processor on each compute node responsive to each processor's simultaneous execution of a message passing computer program instruction of a collective parallel operation, that processor's collective operation contribution to the other processors on that compute node using intra-node communications;transmitting on a designated network link, by each processor on each compute node according to a serial processor transmission sequence, that processor's collective operation contribution to the other processors on the other compute nodes using inter-node communications, wherein the serial processor transmission sequence specifies the order in which each processor on a compute node may access the designated network link for transmitting a collective operation contribution through the network to another compute node,wherein each processor's collective operation contribution comprises data contributed by that processor to effect the collective parallel operation; andwherein transmitting on a designated network link comprises:determining, by each processor on that compute node, that processor's position in the serial processor transmission sequence, andtransferring, by each processor on that compute node, that processor's collective operation contribution to the other processors on the other compute nodes in dependence upon that processor's position in the serial processor transmission sequence. 8. The parallel computer of claim 7 wherein: each compute node has a single designated network link for transmitting the collective operation contributions of each processor on that compute node. 9. The parallel computer of claim 7 wherein: each compute node has a plurality of designated network links for transmitting the collective operation contributions of each processor on that compute node, each designated network link for each compute node uniquely corresponding to one of a plurality of network link domains for that compute node in which a subset of the processors for that compute node are organized, each network link domain having a serial domain processor transmission sequence;the computer memory has disposed within it computer program instructions capable of assigning each processor on each compute node into one of the network link domains for that compute node, including assigning that processor a position in the serial domain processor transmission sequence for that network link domain; andtransmitting on a designated network link, by each processor on each compute node according to a serial processor transmission sequence, that processor's collective operation contribution to the other processors on the other compute nodes using inter-node communications further comprises performing in parallel for each network link domain:determining, by each processor assigned to that network link domain on that compute node, that processor's position in the serial domain processor transmission sequence, andtransferring on the designated network link corresponding to that network link domain, by each processor in that network link domain on that compute node, that processor's collective operation contribution to the other processors on the other compute nodes in dependence upon that processor's position in the serial domain processor transmission sequence. 10. The parallel computer of claim 7 wherein the intra-node communications are carried out using shared memory operations, direct memory access ('DMA') operations, or message passing operations. 11. The parallel computer of claim 7 wherein broadcasting collective operation contributions is carried out as part of a gossiping communications pattern. 12. The parallel computer of claim 7 wherein the plurality of compute nodes are connected together through a plurality of data communications networks, at least one of the data communications networks optimized for point to point data communications, at least one of the data communications networks optimized for collective operations. 13. A computer program product for broadcasting collective operation contributions throughout a parallel computer, the parallel computer comprising a plurality of compute nodes connected together through a data communications network, each compute node having a plurality of processors for use in collective parallel operations on the parallel computer, each collective parallel operation comprising a message passing computer program instruction executed simultaneously by all processors in all compute nodes of the parallel computer, the computer program product disposed upon a computer readable recordable medium, the computer program product comprising computer program instructions capable of: transmitting, by each processor on each compute node responsive to each processor's simultaneous execution of a message passing computer program instruction of a collective parallel operation, that processor's collective operation contribution to the other processors on that compute node using intra-node communications;transmitting on a designated network link, by each processor on each compute node according to a serial processor transmission sequence, that processor's collective operation contribution to the other processors on the other compute nodes using inter-node communications, wherein the serial processor transmission sequence specifies the order in which each processor on a compute node may access the designated network link for transmitting a collective operation contribution through the network to another compute node,wherein each processor's collective operation contribution comprises data contributed by that processor to effect the collective parallel operation; andwherein transmitting on a designated network link comprises:determining, by each processor on that compute node, that processor's position in the serial processor transmission sequence, andtransferring, by each processor on that compute node, that processor's collective operation contribution to the other processors on the other compute nodes in dependence upon that processor's position in the serial processor transmission sequence. 14. The computer program product of claim 13 wherein: each compute node has a single designated network link for transmitting the collective operation contributions of each processor on that compute node. 15. The computer program product of claim 13 wherein: each compute node has a plurality of designated network links for transmitting the collective operation contributions of each processor on that compute node, each designated network link for each compute node uniquely corresponding to one of a plurality of network link domains for that compute node in which a subset of the processors for that compute node are organized, each network link domain having a serial domain processor transmission sequence;the computer program product further comprises computer program instructions capable of assigning each processor on each compute node into one of the network link domains for that compute node, including assigning that processor a position in the serial domain processor transmission sequence for that network link domain; andtransmitting on a designated network link, by each processor on each compute node according to a serial processor transmission sequence, that processor's collective operation contribution to the other processors on the other compute nodes using inter-node communications further comprises performing in parallel for each network link domain:determining, by each processor assigned to that network link domain on that compute node, that processor's position in the serial domain processor transmission sequence, andtransferring on the designated network link corresponding to that network link domain, by each processor in that network link domain on that compute node, that processor's collective operation contribution to the other processors on the other compute nodes in dependence upon that processor's position in the serial domain processor transmission sequence. 16. The computer program product of claim 13 wherein the intra-node communications are carried out using shared memory operations, direct memory access ('DMA') operations, or message passing operations. 17. The computer program product of claim 13 wherein broadcasting collective operation contributions is carried out as part of a gossiping communications pattern. 18. The computer program product of claim 13 wherein the plurality of compute nodes are connected together through a plurality of data communications networks, at least one of the data communications networks optimized for point to point data communications, at least one of the data communications networks optimized for collective operations.