Concurrent, non-blocking, lock-free queue and method, apparatus, and computer program product for implementing same

• US 7,802,032 B2
• Filed: 11/13/2006
• Issued: 09/21/2010
• Est. Priority Date: 11/13/2006
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Christenson, David Alan
• Primary Examiner(s): Sorrell; Eron J
• Application Number:
  US11/559,004
• Publication Number:
  US 20080112423A1
• Time in Patent Office: 1,408 Days
• Field of Search: None
• US Class Current: 710/52
• CPC Class Codes: H04L 49/90 Buffering arrangements
• CPC Class Codes: H04L 49/901 using storage descriptor, e.g. read or write pointers

Abstract

A dummy node is enqueued to a concurrent, non-blocking, lock-free FIFO queue only when necessary to prevent the queue from becoming empty. The dummy node is only enqueued during a dequeue operation and only when the queue contains a single user node during the dequeue operation. This reduces overhead relative to conventional mechanisms that always keep a dummy node in the queue. User nodes are enqueued directly to the queue and can be immediately dequeued on-demand by any thread. Preferably, the enqueueing and dequeueing operations include the use of load-linked/store conditional (LL/SC) synchronization primitives. This solves the ABA problem without requiring the use a unique number, such as a queue-specific number, and contrasts with conventional mechanisms that include the use of compare-and-swap (CAS) synchronization primitives and address the ABA problem through the use of a unique number. In addition, storage ordering fences are preferably inserted to allow the algorithm to run on weakly consistent processors.

Claims

1. A method of implementing a queue, comprising the steps of:

(a) enqueueing a user node to the end of a queue, wherein prior to the enqueueing step (a), the queue comprises at least a dummy node having a next pointer to a next node, and wherein the user node has a next pointer to a next node, and wherein the enqueueing step (a) includes the steps of;

linking the user node to the end of the queue by setting the next pointer of a tail node to point to the user node;

updating a tail pointer of the queue to point to the user node, wherein the updating step is not completed during the enqueueing step (a) but is completed during a subsequent enqueue operation or a subsequent dequeue operation;

(b) after the enqueueing step (a), dequeueing the dummy node from the queue;

(c) after the dequeueing step (b), conditionally dequeueing a head node from the queue;

(d) during the conditional dequeueing step (c), enqueueing another dummy node to the end of the queue but only if the queue contains one and no more than one user node.

2. The method as recited in claim 1, wherein if the conditional dequeueing step (c) is successful in dequeueing a node, the method further comprising the steps of:

setting the next pointer of the node dequeued in step (c) to NULL;

if the node dequeued in step (c) is a user node, returning the dequeued user node;

if the node dequeued in step (c) is a dummy node, restarting the conditional dequeueing step (c).

3. The method as recited in claim 1, wherein the enqueueing step (a), the dequeueing step (b), the conditional dequeueing step (c), and the enqueueing step (d) each includes the use of load-linked/store conditional primitives.

4. The method as recited in claim 3, wherein the enqueueing step (a), the dequeueing step (b), the conditional dequeueing step (c), and the enqueueing step (d) each includes the use of a storage ordering fence.

5. A computer program product for implementing a queue in a digital computing device having at least one processor, comprising:

a plurality of executable instructions provided on computer readable recordable media, wherein the executable instructions, when executed by the at least one processor, cause the digital computing device to perform the steps of;

(a) enqueueing a user node to the end of a queue, wherein prior to the enqueueing step (a), the queue comprises at least a dummy node having a next pointer to a next node, and wherein the user node has a next pointer to a next node, and wherein the enqueueing step (a) includes the steps of;

linking the user node to the end of the queue by setting the next pointer of a tail node to point to the user node;

updating a tail pointer of the queue to point to the user node, wherein the updating step is not completed during the enqueueing step (a) but is completed during a subsequent enqueue operation or a subsequent dequeue operation;

(b) after the enqueueing step (a), dequeueing the dummy node from the queue;

(c) after the dequeueing step (b), conditionally dequeueing a head node from the queue;

(d) during the conditional dequeueing step (c), enqueueing another dummy node to the end of the queue only if the queue contains one and no more than one user node.

6. The computer program product as recited in claim 5, wherein if the conditional dequeueing step (c) is successful in dequeueing a node, the method further comprising the steps of:

setting the next pointer of the node dequeued in step (c) to NULL;

if the node dequeued in step (c) is a user node, returning the dequeued user node;

if the node dequeued in step (c) is a dummy node, restarting the conditional dequeueing step (c).

7. The computer program product as recited in claim 5, wherein the enqueueing step (a), the dequeueing step (b), the conditional dequeueing step (c), and the enqueueing step (d) each includes the use of load-linked/store conditional primitives.

8. The computer program product as recited in claim 7, wherein the enqueueing step (a), the dequeueing step (b), the conditional dequeueing step (c), and the enqueueing step (d) each includes the use of a storage ordering fence.