PERFORMING SYNCHRONIZED COLLECTIVE OPERATIONS OVER MULTIPLE PROCESS GROUPS

• US 20130339506A1
• Filed: 08/31/2012
• Published: 12/19/2013
• Est. Priority Date: 06/13/2012
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: George, Thomas, Jain, Nikhil, Kumar, Sameer, Mittal, Anshul, Sabharwal, Yogish
• Granted Patent: US 9,665,532 B2
• Time in Patent Office: Days
• US Class Current: 709/223
• CPC Class Codes: G06F 15/17318 Parallel communications techniques, e.g. gather, scatter, reduce, roadcast, multicast, all to all

Abstract

Methods and arrangements for performing synchronized collective operations. Communication calls are accepted from at least two distinct processor groups. Edge disjoint spanning paths are created over a collective comprising the processor groups, and the spanning paths are assigned to the processor groups to facilitate communication within each processor group.

Claims

1. A method comprising:

accepting communication calls from at least two distinct processor groups;

creating edge disjoint spanning paths over a collective comprising the processor groups; and

assigning the spanning paths to the processor groups to facilitate communication within each processor group.

2. The method according to claim 1, wherein network architecture is accepted as input.

3. The method according to claim 2, wherein the processor groups are accepted as input.

4. The method according to claim 3, comprising establishing a relationship between the processor groups in accordance with accepting network architecture and the processor groups as input.

5. The method according to claim 1, comprising routing communication for each processor group along at least one assigned spanning path.

6. The method according to claim 1, wherein the spanning paths comprise at least one member selected from the group consisting of:

bandwidth-efficient paths;

load-balancing-efficient paths; and

power-efficient paths.

7. The method according to claim 1, wherein said accepting comprises receiving calls from nodes in the processor groups.

8. The method according to claim 1, wherein said accepting comprises receiving communication calls from two subcommunicators.

9. The method according to claim 8, wherein said creating comprises creating two independent bidirectional spanning trees.

10. The method according to claim 9, wherein said creating of two spanning trees comprises allocating one tree to each of the two subcommunicators.

11. The method according to claim 10, wherein said allocating comprises according a data forwarding function to a spanning tree with respect to nodes of the subcommunicator to which the spanning tree is allocated.

12. The method according to claim 11, wherein said allocating comprises suppressing a data forwarding function to a spanning tree with respect to nodes of the subcommunicator to which the spanning tree is not allocated.

13. The method according to claim 9, wherein the two spanning trees are commonly rooted.

14. A method comprising:

accepting communication calls from nodes, each node belonging to one of at least two distinct processor groups;

communicatively connecting nodes within each processor group;

said connecting comprising;

creating spanning paths over a union of all processor groups;

assigning at least one spanning path to each of the processor groups, to facilitate communication between nodes within each processor group; and

configuring the at least one spanning path of a first one of the processor groups to be edge disjoint with respect to the at least one spanning path of a second one of the processor groups.

15. The method according to claim 14, wherein network architecture is accepted as input.

16. The method according to claim 15, wherein the processor groups are accepted as input.

17. The method according to claim 14, wherein the spanning paths comprise at least one member selected from the group consisting of:

bandwidth-efficient paths;

load-balancing-efficient paths; and

power-efficient paths.

18. The method according to claim 14, wherein:

said creating comprises creating two independent bidirectional spanning trees; and

said assigning comprises assigning each of said spanning trees to a corresponding one of the processor groups.