Techniques for reducing thread overhead for systems with multiple multi-theaded processors

• US 20070283357A1
• Filed: 06/05/2006
• Published: 12/06/2007
• Est. Priority Date: 06/05/2006
• Status: Active Grant

• Current Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Original Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Inventors: Jeter, Robert, Marshall, John, Lee, William, Garner, Trevor
• Granted Patent: US 8,453,147 B2
• Time in Patent Office: Days
• US Class Current: 718/102
• CPC Class Codes: G06F 9/5027 the resource being a machine, e.g. CPUs, Servers, Terminals

Abstract

Techniques for processing requests from a processing thread for a shared resource shared among threads on one or more processors include receiving a bundle of requests from a portion of a thread that is executed during a single wake interval on a particular processor. The bundle includes multiple commands for one or more shared resources. The bundle is processed at the shared resource(s) to produce a bundle result. The bundle result is sent to the particular processor. The thread undergoes no more than one wake interval to sleep interval cycle while the bundle commands are processed at the shared resource(s). These techniques allow a lock for shared resource(s) to be obtained, used and released all while the particular thread is sleeping, so that locks are held for shorter times than in conventional approaches. Using these techniques, line rate packet processing is more readily achieved in routers with multiple multi-threaded processors.

Claims

1. A method for processing requests from a processing thread for a shared resource shared among a plurality of processing threads, comprising the steps of:

receiving, from a portion of a thread that is executed during a single wake interval on a particular processor, a bundle of requests comprising a plurality of commands for one or more shared resources;

processing the bundle of requests at the one or more shared resources to produce a bundle result; and

sending the bundle result to the particular processor,

wherein the thread undergoes no more than one wake interval to sleep interval cycle while the plurality of commands are processed at the one or more shared resources.

2. A method as recited in claim 1, further comprising:

forming the bundle of requests by a thread during a single wake interval of the thread on the particular processor; and

switching threads at the particular processor after forming the bundle of requests.

3. A method as recited in claim 2, further comprising sending each request in the bundle of requests to a distribution process for the one or more shared resources on a bus shared by multiple processors during a time interval reserved for the particular processor.

4. A method as recited in claim 1, said step of sending the bundle result further comprising sending a separate result to the particular processor for each command in the bundle of requests.

5. A method as recited in claim 1, wherein the plurality of commands includes a lock request for a particular portion of a particular resource of the one or more shared resources.

6. A method as recited in claim 4, said step of processing the bundle of requests further comprising:

directing the lock request to a lock controller;

receiving from the lock controller a particular lock for the particular portion of the particular resource; and

after receiving the particular lock, then processing a command that follows the lock request in the bundle of requests.

7. A method as recited in claim 5, wherein the plurality of commands includes a lock release for any lock granted for the particular portion of the particular resource.

8. A method as recited in claim 1, said step of processing the bundle of requests at the one or more shared resources to produce a bundle result further comprising:

retaining a particular result from a command of the plurality of commands at a distribution process for the one or more shared resources; and

executing a subsequent command of the plurality of commands, which subsequent command uses the particular result as input.

9. A method as recited in claim 8, said step of processing the bundle of requests at the one or more shared resources to produce a bundle result further comprising:

determining whether the particular result has been stored in a local register for the distribution process, which register is not in the particular processor; and

if it is determined that the particular result has been stored in the local register, then performing said step of executing the subsequent command.

10. An apparatus for processing requests from a processing thread for a shared resource shared among a plurality of processing threads, comprising:

means for receiving, from a portion of a thread that is executed during a single wake interval on a particular processor, a bundle of requests comprising a plurality of commands for one or more shared resources;

means for processing the bundle of requests at the one or more shared resources to produce a bundle result; and

means for sending the bundle result to the particular processor,

wherein the thread undergoes no more than one wake interval to sleep interval cycle while the plurality of commands are processed at the one or more shared resources.

11. An apparatus for processing requests from a processing thread for a shared resource shared among a plurality of processing threads, comprising:

one or more multi-threaded processors that each allow alternate execution of a plurality of processing threads;

one or more shared resources shared among a plurality of processing threads; and

a shared resource distribution block disposed between the one or more multi-threaded processors and the one or more shared resources,

wherein

the shared resource distribution block is configured to perform the steps of

receiving, from a portion of a particular thread that is executed during a single wake interval on a particular processor, a bundle of requests comprising a plurality of commands for one or more shared resources;

distributing the bundle of requests to the one or more shared resources to produce a bundle result; and

causing the bundle result to be sent to the particular processor while the particular thread undergoes no more than one wake interval to sleep interval cycle.

12. An apparatus as recited in claim 11, further comprising a network interface for communicating a data packet there through.

13. An apparatus as recited in claim 11, further comprising one or more sequences of instructions stored in a computer-readable medium, which, when executed by the particular processor, causes the particular processor to carry out the steps of:

forming the bundle of requests by the particular thread during a single wake interval of the particular thread on the particular processor; and

switching threads at the particular processor after forming the bundle of requests.

14. An apparatus as recited in claim 13, wherein;

the one or more multi-threaded processors include a plurality of processors;

the apparatus further comprises a single bus connecting the plurality of processors to the shared resource distribution block; and

the one or more sequences of instructions further causes the particular processor to carry out the step of sending each req4est in the bundle of requests to the shared resource distribution block during a time interval reserved for the particular processor.

15. An apparatus as recited in claim 11, said step of causing the bundle result to be sent further comprising causing a separate result to be sent to the particular processor for each command in the bundle of requests.

16. An apparatus as recited in claim 11, wherein the plurality of commands includes a lock request for a particular portion of a particular resource of the one or more shared resources.

17. An apparatus as recited in claim 16, said step of distributing the bundle of requests further comprising:

directing the lock request to a lock controller;

receiving from the lock controller a particular lock for the particular portion of the particular resource; and

after receiving the particular lock, then distributing to the particular resource a command that follows the lock request in the bundle of requests.

18. An apparatus as recited in claim 16, wherein the plurality of commands includes a lock release for any lock granted for the particular portion of the particular resource.

19. An apparatus as recited in claim 11, wherein:

the shared resource distribution block comprises a register for storing a result from a command executed by the one or more shared resources; and

said step of distributing the bundle of requests to the one or more shared resources to produce a bundle result further comprising

receiving a particular result from a particular resource of the one or more shared resources in response to a particular command of the plurality of commands distributed to the particular resource,

retaining the particular result in the register,

retrieving the particular result from the register, and

distributing a subsequent command of the plurality of commands, which subsequent command includes the particular result retrieved from the register.

20. An apparatus as recited in claim 19, said step of distributing the bundle of requests at the one or more shared resources to produce a bundle result further comprising:

determining whether the particular result has been stored in the register; and

if it is determined that the particular result has been stored in the register, then performing said steps of retrieving the particular result and distributing the subsequent command.