Managing nodes in a high-performance computing system using a node registrar

• US 9,747,130 B2
• Filed: 06/17/2015
• Issued: 08/29/2017
• Est. Priority Date: 06/16/2011
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Teather, Gregory Wray
• Primary Examiner(s): GHAFFARI, ABU Z
• Application Number:
  US14/741,807
• Publication Number:
  US 20160004563A1
• Time in Patent Office: 804 Days
• Field of Search: None
• CPC Class Codes: G06F 2209/505 Clust
• CPC Class Codes: G06F 9/465 Distributed object oriented systems remote method invocation [RMI] G06F9/548
• CPC Class Codes: G06F 9/4843 by program, e.g. task dispatcher, supervisor, operating system
• CPC Class Codes: G06F 9/4881 Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
• CPC Class Codes: G06F 9/5011 the resources being hardware resources other than CPUs, Servers and Terminals
• CPC Class Codes: G06F 9/5061 Partitioning or combining of resources
• CPC Class Codes: H04L 67/1001 for accessing one among a plurality of replicated servers

Abstract

A method of managing nodes in a high-performance computing (HPC) system, which includes a management subsystem and a job scheduler subsystem, includes providing a node registrar subsystem. Logical node management functions are performed with the node registrar subsystem. Other management functions are performed with the management subsystem using the node registrar subsystem. Job scheduling functions are performed with the job scheduler subsystem using the node registrar subsystem.

Claims

1. A high-performance computing (HPC) system, comprising:

at least one processor;

a memory communicatively coupled to the at least one processor;

a node registrar subsystem, implemented on the at least one processor, that performs logical node management functions and data sharing between HPC subsystems to provide interaction between heterogeneous subsystems, the logical node management functions including at least one of handling state transitions of nodes, adding the nodes, removing the nodes, and updating node properties, wherein the node registrar subsystem comprises a stateless node registrar service and a database for storing node information for nodes in the HPC system, the stateless node registrar service providing high-availability through scale-out of multiple services rather than relying on failover;

a management subsystem, implemented on the at least one processor, that performs other management functions other than the logical node management functions, wherein the management subsystem communicates with the node registrar subsystem in performing the other management functions; and

a job scheduler subsystem, implemented on the at least one processor, that performs job scheduling functions, wherein the job scheduler subsystem communicates with the node registrar subsystem in performing the job scheduling functions, wherein the management subsystem, the job scheduler subsystem, and the node registrar subsystem are separate, heterogeneous subsystems, and wherein the node registrar subsystem provides communication connections that allow direct communication interaction between the management subsystem, the job scheduler subsystem, and the node registrar subsystem.

2. The HPC system of claim 1, wherein the management subsystem and the job scheduler subsystem access the database using the stateless node registrar service of the node registrar subsystem.

3. The HPC system of claim 1, wherein the management subsystem communicates with the node registrar service to access the stored node information.

4. The HPC system of claim 1, wherein the management subsystem communicates with the stateless node registrar service to update node properties in the database.

5. The HPC system of claim 1, wherein the management subsystem communicates with the stateless node registrar service to query nodes by property.

6. The HPC system of claim 1, wherein the management subsystem communicates with the stateless node registrar service to query nodes by group.

7. The HPC system of claim 1, wherein the job scheduler subsystem communicates with the stateless node registrar service to access the stored node information.

8. The HPC system of claim 1, wherein the job scheduler subsystem communicates with the stateless node registrar service to update node properties.

9. The HPC system of claim 1, wherein the job scheduler subsystem communicates with the stateless node registrar service to query nodes by property.

10. The HPC system of claim 1, wherein the job scheduler subsystem communicates with the stateless node registrar service to query nodes by group.

11. The HPC system of claim 1, wherein the database includes a node table with individual rows in the node table corresponding to individual nodes in the HPC system, and individual columns listing properties of the individual nodes in the HPC system.

12. The HPC system of claim 1, wherein the management subsystem and the job scheduler subsystem are each a client of the node registrar subsystem.

13. The HPC system of claim 1, wherein the other management functions performed by the management subsystem include at least one of node deployment, node configuration management, infrastructure configuration management, or node monitoring.

14. A node registrar subsystem for a high-performance computing (HPC) system, the node registrar subsystem comprising:

at least one processor;

a memory communicatively coupled to the at least on processor;

a database, implemented on the memory, that stores node information for nodes in the HPC system; and

a stateless node registrar service, implemented on the at least one processor, that performs logical node management functions, communicates with a management subsystem to facilitate access to the database by the management subsystem in response to the management subsystem performing other management functions other than the logical node management functions, and communicates with a job scheduler subsystem to facilitate access to the database by the job scheduler subsystem in response to the job scheduler subsystem performing job scheduling functions, the stateless node registrar service providing high-availability through scale-out of multiple services rather than relying on failover, the logical node management functions including at least one of handling state transitions of the nodes, adding the nodes, removing the nodes, and updating node properties, wherein the management subsystem and the job scheduler subsystem are separate heterogeneous subsystems from the node registrar subsystem, and wherein the node registrar subsystem provides communication connections between the management subsystem, the job scheduler subsystem, and the node registrar subsystem.

15. The node registrar subsystem of claim 14, wherein the other management functions performed by the management subsystem include at least one of node deployment, node configuration management, infrastructure configuration management, or node monitoring.

16. The node registrar subsystem of claim 14, wherein the management subsystem and the job scheduler subsystem access the database of the node registrar subsystem using the stateless node registrar service.

17. The node registrar subsystem of claim 14, wherein the database includes a node table with individual rows in the node table corresponding to individual nodes in the HPC system, and individual columns listing properties of the individual nodes in the HPC system.

18. A method of managing nodes in a high-performance computing (HPC) system, the method comprising:

performing, by a node registrar subsystem implemented on at least one processor, logical node management functions in the HPC system, the logical node management functions including at least one of handling state transitions of nodes, adding the nodes, removing the nodes, and updating node properties, wherein the node registrar subsystem comprises a stateless node registrar service and a database for storing node information for the nodes in the HPC system, the stateless node registrar service providing high-availability through scale-out of multiple services rather than relying on failover;

providing, by the node registrar subsystem, communication connections between a management subsystem, a job scheduler subsystem, and the node registrar subsystem, the management subsystem and the job scheduler subsystem being separate heterogeneous subsystems from the node registrar subsystem;

performing other management functions other than the logical node management functions to be performed within the HPC system with the management subsystem, the management subsystem directly communicating with the node registrar subsystem via the communication connections provided by the node registrar subsystem to perform the other management functions; and

performing job scheduling functions within the HPC system with the job scheduler subsystem, the job scheduler subsystem directly communicating with the node registrar subsystem via the communication connections to perform the job scheduling functions.

19. The method of claim 18, further comprising:

providing the management subsystem and the job scheduler subsystem access to the database of the node registrar subsystem using the stateless node registrar service.

20. The method of claim 18, wherein the database includes a node table with individual rows in the node table corresponding to individual nodes in the HPC system, and individual columns listing properties of the individual nodes in the HPC system.