Granted Patent B2

US 12,554,668 · App. 18/748,324 · Granted Feb 17, 2026

Inventor: Rujie Chen (Shenzhen, CN)

Assignee: HUAWEI TECHNOLOGIES CO., LTD.

G06F13/4068

Abstract

Claims (77)

1 . An apparatus comprising:

N computing nodes;

a device manager configured to:

obtain N interface informations of the N computing nodes; and

determine first interconnection relationships among the N computing nodes based on the N interface informations; and

a system manager configured to group, based on the first interconnection relationships and N computing power informations of the N computing nodes, the N computing nodes into M collaboration groups,

wherein N is an integer greater than 1,

wherein M is an integer greater than 0,

wherein each of the M collaboration groups comprises at least one computing node of the N computing nodes, and

wherein within each of the M collaboration groups all of the at least one computing node have a same computing power and are physically interconnected.

2 . The apparatus of claim 1 , further comprising L chips, wherein the L chips comprise the N computing nodes, wherein the device manager comprises L device sub-managers respectively in the L chips, wherein L is an integer greater than 0 and less than N, wherein the L device sub-managers comprise a first device sub-manager in a first chip of the L chips, wherein the first chip comprises K computing nodes of the N computing nodes, and wherein the first device sub-manager is configured to:

obtain K interface informations that are of the K computing nodes and that are in the N interface informations; and

determine a second interconnection relationship among the K computing nodes based on the K interface informations.

3 . The apparatus of claim 1 , further comprising:

a chip board comprising the device manager; and

N chips comprising the N computing nodes and cascaded on the chip board.

4 . The apparatus of claim 1 , wherein the system manager is further configured to:

obtain pre-grouping information by pre-grouping the N computing nodes into pre-groups, wherein within each of the pre-groups all of a plurality of computing nodes of the N computing nodes are capable of running a same application; and

send the pre-grouping information to the device manager, and

wherein the device manager is further configured to determine the first interconnection relationships based on the pre-grouping information and the N interface informations.

5 . The apparatus of claim 1 , wherein the device manager is further configured to:

obtain an identification information table from a target computing node of the N computing nodes; and

associate, based on the identification information table, interconnected computing nodes of the N computing nodes.

6 . The apparatus of claim 5 , wherein the N computing nodes have N physical device identifiers and are configured to:

generate mark identifiers; and

establish mapping relationships between the mark identifiers and the N physical device identifiers.

7 . The apparatus of claim 6 , wherein the system manager is further configured to:

obtain the mapping relationships; and

generate the identification information table, wherein the identification information table comprises the mapping relationships.

8 . The apparatus of claim 6 , wherein the device manager is further configured to:

obtain the identification information table from the target computing node based on a first physical device identifier that is of the target computing node and that is in the N physical device identifiers; and

associate, based on the identification information table, the mark identifiers of the interconnected computing nodes.

9 . The apparatus of claim 1 , wherein the system manager is further configured to group interconnected computing nodes that are of the N computing nodes, have a same computing power, and have a same function into a first collaboration group of the M collaboration groups.

10 . The apparatus of claim 6 , wherein the system manager is further configured to:

select a target collaboration group from the M collaboration groups based on a computing power requirement required by a to-be-executed task, wherein the to-be-executed task comprises parallel tasks, and wherein the target collaboration group comprises first computing nodes of the N computing nodes; and

respectively allocate the parallel tasks to of the first computing nodes based on mark identifiers of the first computing nodes.

11 . The apparatus of claim 10 , wherein the device manager is further configured to:

determine, based on the identification information table, first physical device identifiers that are of the N physical device identifiers and that correspond to the mark identifiers; and

respectively allocate, based on the first physical device identifiers, the parallel tasks to the first computing nodes.

12 . A method comprising:

obtaining N interface informations of N computing nodes;

determining first interconnection relationships among the N computing nodes based on the N interface informations; and

grouping, based on the first interconnection relationships and N computing power informations of the N computing nodes, the N computing nodes into M collaboration groups,

wherein N is an integer greater than 1,

wherein M is an integer greater than 0,

wherein each of the M collaboration groups comprises at least one computing node of the N computing nodes, and

wherein within each of the M collaboration groups all of the at least one computing node have a computing power and are physically interconnected.

13 . The method of claim 12 , wherein the N computing nodes comprise K computing nodes in a chip, and wherein the method further comprises:

obtaining K interface informations that are of the K computing nodes and that are in the N interface informations; and

determining a second interconnection relationship among the K computing nodes based on the K interface informations.

14 . The method of claim 12 , further comprising:

obtaining pre-grouping information by pre-grouping the N computing nodes into pre-groups, wherein within each of the pre-groups a plurality of computing nodes of the N computing nodes are capable of running a same application;

sending the pre-grouping information to a device manager; and

determining the first interconnection relationships based on the pre-grouping information and the N interface informations.

15 . The method of claim 12 , further comprising:

obtaining an identification information table from a target computing node of the N computing nodes; and

associating, based on the identification information table, interconnected computing nodes of the N computing nodes.

16 . The method of claim 15 , further comprising:

generating mark identifiers; and

establishing mapping relationships between the mark identifiers and N physical device identifiers of the N computing nodes.

17 . The method of claim 16 , further comprising:

obtaining the mapping relationships; and

generating the identification information table, wherein the identification information table comprises the mapping relationships.

18 . A computer program product comprising instructions that are stored on a non-transitory computer-readable medium and that, when executed by one or more processors, cause an apparatus to:

obtain N interface informations of N computing nodes;

determine an interconnection relationship among the N computing nodes based on the N interface informations; and

group, based on the interconnection relationship and computing power information of the N computing nodes, the N computing nodes into M collaboration groups,

wherein N is an integer greater than 1,

wherein M is an integer greater than 0,

wherein each of the M collaboration groups comprises at least one computing node of the N computing nodes, and

wherein within each of the M collaboration groups all of the at least one computing node have a same computing power and are physically interconnected.

19 . The method of claim 16 , further comprising:

obtaining the identification information table from the target computing node based on a first physical device identifier that is of the target computing node and that is in the N physical device identifiers; and

associating, based on the identification information table, the mark identifiers of the interconnected computing nodes.

20 . The method of claim 16 , further comprising:

selecting a target collaboration group from the M collaboration groups based on a computing power requirement required by a to-be-executed task, wherein the to-be-executed task comprises parallel tasks, and wherein the target collaboration group comprises first computing nodes of the N computing nodes; and

respectively allocating the parallel tasks to the first computing nodes based on mark identifiers of the first computing nodes.

Assignments (1)

Continuity (2)

References Cited (14)