Resource management in a clustered computer system

• US 6,338,112 B1
• Filed: 05/18/2000
• Issued: 01/08/2002
• Est. Priority Date: 02/21/1997
• Status: Expired due to Term

• Current Assignee: RPX Corporation
• Original Assignee: Novell Incorporated (Open Text Corporation)
• Inventors: Wipfel, Robert A, Murphy, David
• Primary Examiner(s): Dharia, Rupal
• Application Number:
  US09/574,093
• Time in Patent Office: 600 Days
• Field of Search: 709/223, 709/224, 709/226, 709/251, 710/260, 710/200, 710/269
• US Class Current: 710/269
• CPC Class Codes: G06F 11/008 Reliability or availability analysis

Abstract

Methods, systems, and devices are provided for managing resources in a computing cluster. The managed resources include cluster nodes themselves, as well as sharable resources such as memory buffers and bandwidth credits that may be used by one or more nodes. Resource management includes detecting failures and possible failures by node software, node hardware, interconnects, and system area network switches and taking steps to compensate for failures and prevent problems such as uncoordinated access to a shared disk. Resource management also includes reallocating sharable resources in response to node failure, demands by application programs, or other events. Specific examples provided include failure detection by remote memory probes, emergency communication through a shared disk, and sharable resource allocation with minimal locking.

Claims

1. A method for managing resources in a cluster, the method comprising the computer-implemented steps of:

unlocking a global queue of resources which is guarded by a lock;

updating the unlocked global queue;

locking the updated global queue; and

updating a local queue of resources while executing an interrupt handler.

2. The method of claim 1, wherein the step of updating the local queue occurs before the step of unlocking the global queue.

3. The method of claim 1, wherein the step of updating the local queue occurs after the step of locking the updated global queue.

4. The method of claim 1, wherein the updating steps are preceded by the step of determining that reliable communication with a cluster node over a system area network has failed, and the updating steps assist in returning to the global queue resources that were previously in the local queue on the node.

5. A computer system comprising:

at least two interconnected nodes capable of presenting a uniform system image such that an application program views the interconnected nodes as a single computing platform; and

a management means for managing computational resources for use by the nodes, wherein the management means comprises a queue and lock management means for managing access to a global and local groups of sharable resources using a single lock and at least one interrupt handler.

6. The system of claim 5, wherein the queue and lock management means comprises an interrupt disabling means for saving the state of a single processor and disabling at least one interrupt to the processor, and also comprises a restoring means for restoring the state of the processor.

7. The system of claim 5, wherein the lock is a mutex.

8. The system of claim 5, wherein the queue and lock management means comprises means for moving a resource from the global group of sharable resources to a local group of sharable resources.

9. The system of claim 8, wherein the global group of sharable resources is a global queue.

10. The system of claim 8, wherein the global group of sharable resources is a global tree.

11. The system of claim 8, wherein the sharable resources include memory buffers.

12. The system of claim 8, wherein the sharable resources include bandwidth credits.

13. The system of claim 8, wherein at least one local group of sharable resources maintains resources for use by code running on any of a group of at least two processors in a processor set.

14. The system of claim 5, wherein the queue and lock management means comprises means for making room for more resources in a local group of sharable resources.

15. The system of claim 5, wherein the queue and lock management means comprises means for redistributing sharable resources among at least two local groups.

16. The system of claim 5, wherein at least one of the nodes is a special-purpose raphics node.

17. The system of claim 5, wherein at least one of the nodes is a special-purpose signal processing node.

19. The system of claim 5, wherein the management means further comprises a means for detecting node failure by remotely probing memory.

20. The system of claim 5, wherein the management means further comprises a means for detecting interconnect failure by remotely probing memory.

18. A computer storage medium having a configuration that represents data and instructions which will cause at least a portion of a computer system to perform method steps for managing resources in a cluster computing system, the method steps comprising the steps of unlocking a global queue of resources which is guarded by a lock, updating the unlocked global queue, locking the updated global queue, and updating a local queue of resources while executing an interrupt handler.