PREVENTING MESSAGING QUEUE DEADLOCKS IN A DMA ENVIRONMENT

• US 20110173287A1
• Filed: 09/30/2008
• Published: 07/14/2011
• Est. Priority Date: 09/30/2008
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Gooding, Thomas, Parker, Jeff, Chen, Dong, Blocksome, Michael A., Heidelberger, Philip
• Granted Patent: US 8,631,086 B2
• Time in Patent Office: Days
• US Class Current: 709/212
• CPC Class Codes: G06F 15/17331 Distributed shared memory [DSM], e.g. remote direct memory access [RDMA]

Abstract

Embodiments of the invention may be used to manage message queues in a parallel computing environment to prevent message queue deadlock. A direct memory access controller of a compute node may determine when a messaging queue is full. In response, the DMA may generate an interrupt. An interrupt handler may stop the DMA and swap all descriptors from the full messaging queue into a larger queue (or enlarge the original queue). The interrupt handler then restarts the DMA. Alternatively, the interrupt handler stops the DMA, allocates a memory block to hold queue data, and then moves descriptors from the full messaging queue into the allocated memory block. The interrupt handler then restarts the DMA. During a normal messaging advance cycle, a messaging manager attempts to inject the descriptors in the memory block into other messaging queues until the descriptors have all been processed.

Claims

1. A method for managing message queues in a parallel computing system having a plurality of compute nodes, comprising:

determining that a first queue, on a first compute node, storing a set of message descriptors has become full, wherein a direct memory access controller (DMA) is configured to inject message descriptors into the first queue; and

generating an interrupt delivered to an interrupt handler, wherein the interrupt handler is configured to perform the steps of;

stopping the DMA controller;

generating a second queue, wherein the second queue is larger than the first queue;

swapping the first queue with the second queue such that the DMA controller is configured to inject message descriptors into the second queue;

copying the set of message descriptors from the first queue to the second queue; and

restarting the DMA controller.

2. The method of claim 1, wherein the message descriptors are sent to the first compute node from a second compute node over a network connection connecting the first compute node and the second compute node.

3. The method of claim 1, further comprising, freeing memory allocated to the first queue.

4. The method of claim 1, wherein the message descriptors are generated in response to a remote get packet sent to the first compute node by a second compute node.

5. The method of claim 4, wherein the message descriptors identify a location and a size of data requested to be sent from the first compute node to the second compute node.

6. The method of claim 5, wherein the DMA controller is configured to perform the steps of:

retrieving a message descriptor from the second queue;

retrieving the data at the location identified in the retrieved message descriptor; and

transmitting the data to the second compute node.

7. The method of claim 4, wherein the remote get packet identifies a message descriptor to be injected into the first queue.

8. The method of claim 1, further comprising, after restarting the DMA controller:

receiving, from a second compute node, a remote get packet which includes a message descriptor to be injected into the first queue; and

injecting the message descriptor into the first queue.

9. The method of claim 1, wherein generating a second queue comprises enlarging a memory region allocated to the first queue.

10. A computer-readable storage-medium containing a program which, when executed, performs an operation for managing message queues in a parallel computing system having a plurality of compute nodes, the operation comprising:

determining that a first queue, on a first compute node, storing a set of message descriptors has become full, wherein a direct memory access controller (DMA) is configured to inject message descriptors into the first queue; and

generating an interrupt delivered to an interrupt handler, wherein the interrupt handler is configured to perform the steps of;

stopping the DMA controller;

generating a second queue, wherein the second queue is larger than the first queue;

swapping the first queue with the second queue such that the DMA controller is configured to inject message descriptors into the second queue;

copying the set of message descriptors from the first queue to the second queue; and

restarting the DMA controller.

11. The computer-readable storage-medium of claim 10, wherein the message descriptors are sent to the first compute node from a second compute node over a network connection connecting the first compute node and the second compute node.

12. The computer-readable storage-medium of claim 10, wherein the interrupt handler is further configured to free memory allocated to the first queue.

13. The computer-readable storage-medium of claim 10, wherein the message descriptors are generated in response to a remote get packet sent to the first compute node by a second compute node.

14. The computer-readable storage-medium of claim 13, wherein the message descriptors identify a location and a size of data requested to be sent from the first compute node to the second compute node.

15. The computer-readable storage-medium of claim 14, wherein the DMA controller is configured to perform the steps of:

retrieving a message descriptor from the second queue;

retrieving the data at the location identified in the retrieved message descriptor; and

transmitting the data to the second compute node.

16. The computer-readable storage-medium of claim 14, wherein the remote get packet identifies a message descriptor to be injected into the first queue.

17. The computer-readable storage-medium of claim 10, wherein the operation further comprises:

after restarting the DMA controller, receiving, from a second compute node, a remote get packet which includes a message descriptor to be injected into the first queue; and

injecting the message descriptor into the first queue.

18. The computer-readable storage-medium of claim 10, wherein generating a second queue comprises enlarging a memory region allocated to the first queue.

19. A parallel computing system, comprising:

a plurality of compute nodes, each having at least a processor, a memory and a direct memory access controller (DMA), wherein the plurality of compute nodes are configured to move messages between two compute nodes of the plurality, and wherein the DMA on a first compute node is configured to;

determine that a first queue, on the first compute node, storing a set of message descriptors has become full, and

generate an interrupt delivered to an interrupt handler on the first compute node, wherein the interrupt handler is configured to perform the steps of;

stopping the DMA controller;

generating a second queue, wherein the second queue is larger than the first queue;

swapping the first queue with the second queue such that the DMA controller is configured to inject message descriptors into the second queue;

copying the set of message descriptors from the first queue to the second queue; and

restarting the DMA controller.

20. The parallel computing system of claim 19, wherein the interrupt handler is further configured to free memory allocated to the first queue.

21. The parallel computing system of claim 19, wherein the message descriptors are generated in response to a remote get packet sent to the first compute node by a second compute node.

22. The parallel computing system of claim 21, wherein the message descriptors identify a location and a size of data requested to be sent from the first compute node to the second compute node.

23. The parallel computing system of claim 22, wherein the DMA controller is configured to perform the steps of:

retrieving a message descriptor from the second queue;

retrieving the data at the location identified in the retrieved message descriptor; and

transmitting the data to the second compute node.

24. The parallel computing system of claim 19, wherein the remote get packet identifies a message descriptor to be injected into the first queue.