IO completion architecture for user-mode networking

• US 20030204552A1
• Filed: 04/30/2002
• Published: 10/30/2003
• Est. Priority Date: 04/30/2002
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Zuberi, Khawar M.
• Granted Patent: US 6,988,268 B2
• Time in Patent Office: Days
• US Class Current: 718/103
• CPC Class Codes: H04L 69/12 Protocol engines

Abstract

A new method and framework for implementing network protocol processing utilizing a combination of application threads and a dedicated thread to process IO completions in a completion queue that automatically detects and adjusts thread priorities to alleviate manual intervention. According to the present invention, as data transfer operations are completed by the network interface, completion information identifying the data transfer operations is posted on the completion queue. The completion information is read and processed by a combination of application and dedicated threads running in the system. A method monitors performance of the system to detect whether poor processor utilization or excessive context switches occurs, in which case a different thread is used to process the completion information. In order to context switch to a different thread, the priority level of the dedicated thread is set to a HIGH level to utilize the dedicated thread or the priority level of the dedicated thread is set a LOW level to utilize an application thread.

Claims

1. A method for adaptively processing completions of data transfer operations in a server comprising a network interface communicably coupled to perform the data transfer operations, a completion queue storing completion information for the data transfer operations, a set of executing threads capable of processing the completion information, the set of executing threads comprising a dedicated thread having a priority level responsive to a status of the completion queue, and wherein an application thread is coupled to an application running on the server to process application system calls, the method comprising the steps of:

first processing, by the set of executing threads, the completion information in the completion queue according to a first priority level of the dedicated thread;

determining a change in status of the completion queue and in response assigning a second priority level to the dedicated thread; and

second processing, by the set of executing threads, the completion queue in the completion queue according to the second priority level of the dedicated thread.

2. The method of claim 1 wherein the set of executing threads further comprises the application thread.

3. The method of claim 2 wherein the completion information comprises one or more entries in the completion queue, each entry uniquely identifying a data transfer operation completed by the network interface.

4. The method of claim 3 wherein the first processing step comprises the steps of:

reading the entries in the completion queue;

increasing a counter by one for each entry read, the counter tracking the number of entries processed by the executing threads; and

processing the entries.

5. The method of claim 2 wherein the application thread includes a priority level, the method further comprising the step of determining a base priority level for the application thread.

6. The method of claim 5 wherein the step of determining the base priority level comprises the steps of:

collecting a sample set of priority levels for the application thread invoking system calls; and

analyzing the sample set to determine the base priority level.

7. The method of claim 6 wherein the system calls comprise send calls invoked by the application.

8. The method of claim 6 wherein the step of analyzing the sample set comprises the conditionally executed steps of:

comparing the priority levels in the sample set to each other;

setting the base priority level to a level equal to one of the priority levels in the sample set if the priority levels in the sample set are identical; and

otherwise setting the base priority level to the level equal to a percentile of the sample set if the priority levels in the sample set are not identical.

9. The method of claim 8 wherein the percentile of the sample represents the largest percentile of identical priority levels in the sample set.

10. The method of claim 3 wherein a set of potential statuses for the completion queue include:

an empty completion queue, and the completion queue includes a number of pending entries that exceeds a specified number and wherein the determining a change in status step comprises the steps of;

discontinuing processing the completion information in the completion queue; and

assigning the second priority level to the dedicated thread;

11. The method of claim 10, wherein processing of completion entries during the first processing is performed solely by the dedicated thread and processing of completion entries during the second processing is performed by the application thread without assistance from the dedicated thread, wherein assigning the second priority level step comprises the steps of:

setting the priority level of the dedicated thread to the base priority level; and

comparing the priority level of the dedicated thread to the priority level of the application thread and in response to determining that the application thread priority level is higher than the dedicated thread priority level context switching to the application thread.

12. The method of claim 10 wherein processing of completion entries during the first processing is performed solely by the application thread and processing of completion entries during the second processing is performed by the dedicated thread without assistance from the application thread, wherein assigning the second priority level step comprises the steps of:

setting the priority level of the dedicated thread to a high priority level; and

comparing the priority level of the dedicated thread to the priority level of the application thread and in response to determining that the dedicated thread priority level is higher than the application thread priority level context switching to the dedicated thread.

13. The method of claim 12 wherein the high priority level is higher than the base priority level.

14. The method of claim 13 wherein the high priority level is equal to THREAD_PRIORITY_TIME_CRITICAL.

15. A computer-readable medium having computer-executable instructions for facilitating adaptively processing completions of data transfer operations in a server comprising a network interface communicably coupled to perform the data transfer operations, a completion queue storing completion information for the data transfer operations, a set of executing threads capable of processing the completion information, the set of executing threads comprising a dedicated thread having a priority level responsive to a status of the completion queue, and wherein an application thread is coupled to an application running on the server to process application system calls, the computer-readable medium having computer-executable instructions facilitating performing the steps of:

first processing, by the set of executing threads, the completion information in the completion queue according to a first priority level of the dedicated thread;

determining a change in status of the completion queue and in response assigning a second priority level to the dedicated thread; and

second processing, by the set of executing threads, the completion queue in the completion queue according to the second priority level of the dedicated thread.

16. The computer-readable medium of claim 15 wherein the set of executing threads further comprises the application thread.

17. The computer-readable medium of claim 16 wherein the completion information comprises one or more entries in the completion queue, each entry uniquely identifying the data transfer operation completed by the network interface.

18. The computer-readable medium of claim 17 comprising further computer-executable instructions facilitating performing the steps of:

reading entries in the completion queue;

increasing a counter by one for each entry read, the counter tracking the number of entries processed by the executing threads; and

processing the entries.

19. The computer-readable medium of claim 16 wherein the application thread includes a priority level, the computer-readable medium comprising further computer-executable instructions facilitating performing the step of determining a base priority level for the application thread.

20. The computer-readable medium of claim 19 wherein the computer-executable instructions facilitating performing the step of determining the base priority level comprises computer-executable instructions facilitating performing the steps of:

collecting a sample set of priority levels for the application thread invoking system calls; and

analyzing the sample set to determine the base priority level.

21. The computer-readable medium of claim 20 wherein the system calls comprise send calls invoked by the application.

22. The computer-readable medium of claim 20 wherein the computer-executable instructions facilitating performing the step of analyzing the sample set comprises computer-executable instructions facilitating performing the conditionally executed steps of:

comparing the priority levels in the sample set to each other;

setting the base priority level to a level equal to one of the priority levels in the sample set if the priority levels in the sample set are identical; and

otherwise setting the base priority level to the level equal to a percentile of the sample set if the priority levels in the sample set are not identical.

23. The computer-readable medium of claim 22 wherein the percentile of the sample represents the largest percentile of identical priority levels in the sample set.

24. The computer-readable medium of claim 17 wherein a set of potential statuses for the completion queue include:

an empty completion queue, and the completion queue includes a number of pending entries that exceeds a specified number and wherein the computer-executable instructions facilitating determining a change in status comprises computer-executable instructions facilitating performing the steps of;

discontinuing processing the completion information in the completion queue; and

assigning the second priority level to the dedicated thread;

25. The computer-readable medium of claim 24, wherein processing of completion entries during the first processing is performed solely by the dedicated thread and processing of completion entries during the second processing is performed by the application thread without assistance from the dedicated thread, wherein the computer-executable instructions facilitating assigning the second priority level comprises computer-executable instructions facilitating performing the steps of:

setting the priority level of the dedicated thread to the base priority level; and

comparing the priority level of the dedicated thread to the priority level of the application thread and in response to determining that the application thread priority level is higher than the dedicated thread priority level context switching to the application thread.

26. The computer-readable medium of claim 24 wherein processing of completion entries during the first processing is performed solely by the application thread and processing of completion entries during the second processing is performed by the dedicated thread without assistance from the application thread, wherein the computer-executable instructions facilitating assigning the second priority level step comprises computer-executable instructions facilitating performing the steps of:

setting the priority level of the dedicated thread to a high priority level; and

comparing the priority level of the dedicated thread to the priority level of the application thread and in response to determining that the dedicated thread priority level is higher than the application thread priority level context switching to the dedicated thread.

27. The computer-readable medium of claim 26 wherein the high priority level is higher than the base priority level.

28. The computer-readable medium of claim 27 wherein the high priority level is equal to THREAD_PRIORITY_TIME_CRITICAL.

29. An IO completion framework in a computer system for adaptively processing completion entries in a completion queue, the framework comprising:

a network interface for receiving and enqueueing the completion entries in the completion queue; and

a set of executing threads comprising;

at least a dedicated thread, the dedicated thread having a priority attribute; and

a monitor for setting the priority attribute in accordance with a status of the completion queue.

30. The framework in claim 29 wherein the executing threads include at least one application thread.

31. The framework of claim 30 wherein each completion entry uniquely identifies the data transfer operation completed by the network interface.

32. The framework of claim 30 wherein the monitor comprises:

a counter for tabulating the number of completion entries processed by the executing thread;

an indicator signaling that the completion queue is empty;

a base priority level indicative of priority levels for the system calls invoked by the application; and

a high priority level higher than the base priority level.

33. The framework of claim 30 wherein the statuses of the completion queue include:

an empty completion queue; and

the completion queue includes a number of pending entries that exceeds a specified number.