Load balancing algorithm for servicing client requests

• US 8,707,323 B2
• Filed: 12/30/2005
• Issued: 04/22/2014
• Est. Priority Date: 12/30/2005
• Status: Active Grant

• Current Assignee: SAP SE
• Original Assignee: SAP AG (SAP SE)
• Inventors: Galchev, Galin, Fleischer, Christian, Luik, Oliver
• Primary Examiner(s): KESSLER, GREGORY AARON
• Application Number:
  US11/322,399
• Publication Number:
  US 20070156869A1
• Time in Patent Office: 3,035 Days
• Field of Search: 718/105
• US Class Current: 718/105
• CPC Class Codes: G06F 9/505 considering the load

Abstract

A connection manager includes a dispatcher that dispatches requests to worker nodes in an application server according to a load balancing algorithm.

Claims

1. An application server comprising:

a connection manager to receive a request associated with an existing distributable session;

a request notification queue communicatively coupled to the connection manager to store a request notification of the request, the request notification to be retrieved by a worker node assigned to process the request, wherein the request notification queue is one of a plurality of request notification queues, the worker node is one of a plurality of worker nodes, and each one of the plurality of request notification queues is associated with a respective one of the plurality of worker nodes;

a memory communicatively coupled to the connection manager to store request data, response data and session data of the request; and

a dispatcher communicatively coupled to the connection manager to send the request to the worker node assigned to handle the session if the worker node is available, the dispatcher identifying the worker node assigned to handle the session by querying the session data to determine if the session is already existing and retrieving the worker node assigned to handle the session, the worker node processing the request by retrieving the request notification from the request notification queue associated with the worker node, and the request data and the session data from the memory, the dispatcher to select another worker node from the plurality of worker nodes to which to send the request according to a load balancing algorithm if the worker node is not available.

2. The application server of claim 1, wherein the worker node is available if there are no active requests for the session that the worker node assigned to handle the session has yet to process.

3. The application server of claim 2, wherein the worker node is available if additionally there are no active request notifications pending in the request notification queue associated with the worker node assigned to handle the session.

4. The application server of claim 1, wherein the worker node is not available if the worker node is busy.

5. The application server of claim 1, wherein the dispatcher to select the other worker node to which to send the request comprises the dispatcher to select on a round-robin basis the other worker node from a list of worker nodes maintained in a data structure accessible to the dispatcher.

6. The application server of claim 1, wherein the dispatcher to select another worker node to which to send the request comprises the dispatcher to select on a weighted basis the other worker node from a list of worker nodes maintained in a data structure accessible to the dispatcher.

7. The application server of claim 6, wherein the dispatcher selects the other worker node based on a lowest number of request notifications in the request notification queue associated with the other worker node as compared to a number of request notifications in a respective request notification queue associated with each of the worker nodes in the list.

8. The application server of claim 7, wherein the dispatcher traverses the list of worker nodes and selects a first worker node with an empty associated request notification queue.

9. The application server of claim 8, wherein if there is a tie among two or more worker nodes in the list based on the number of request notifications in the respective request notification queues, the dispatcher selects the worker node that has at least one idle thread.

10. A method comprising:

receiving a request associated with an existing distributable session at a connection manager;

receiving a request notification of the request at a request notification queue from the connection manager, the request notification being retrieved by a worker node assigned to process the request, wherein the request notification queue is one of a plurality of request notification queues, the worker node is one of a plurality of worker nodes, and each one of the plurality of request notification queues is associated with a respective one of the plurality of worker nodes;

storing request data, response data and session data of the request in a memory, the memory being accessible to the connection manager and the worker node;

examining at a dispatcher whether the worker node assigned to handle the session is available to process the request, the examining including querying the session data to determine if the session is an existing session and retrieving the worker node assigned to handle the session;

sending the request from the dispatcher to the assigned worker node if the assigned worker node is available, the worker node processing the request by retrieving the request notification from the request notification queue associated with the worker node and the request data and the session data from the memory; and

selecting another worker node from the plurality of worker nodes to which to send the request according to a load balancing algorithm if the assigned worker node is not available.

11. The method of claim 10 further comprising determining that the assigned worker node is available if there are no active requests for the session that the assigned worker node has yet to process.

12. The method of claim 11, wherein the assigned worker node is available if there are no active request notifications pending in the request notification queue associated with the assigned worker node.

13. The method of claim 10, wherein selecting another worker node to which to send the request comprises selecting on a round-robin basis the other worker node from a list of worker nodes maintained in a data structure accessible to the dispatcher.

14. The method of claim 10, wherein selecting another worker node to which to send the request comprises selecting on a weighted basis the other worker node from a list of worker nodes maintained in a data structure accessible to the dispatcher.

15. The method of claim 14, wherein selecting the other worker node is based on a lowest number of request notifications in the request notification queue associated with the other worker node as compared to a number of request notifications in a respective request notification queue associated with each of the worker nodes in the list.

16. The method of claim 14, wherein selecting the other worker node comprises traversing by the dispatcher the list of worker nodes and selecting a first worker node with an empty associated request notification queue.

17. The method of claim 16, wherein selecting the other worker node comprises selecting the worker node that has at least one idle thread if there is a tie among two or more worker nodes in the list based on the number of request notifications in respective request notification queues.

18. An article of manufacture, comprising:

an electronically accessible medium comprising instructions, that when executed by a processor cause the processor to;

receive a request associated with an existing distributable session at a connection manager;

receive a request notification of the request at a request notification queue from the connection manager, the request notification being retrieved by a worker node assigned to process the request, wherein the request notification queue is one of a plurality of request notification queues, the worker node is one of a plurality of worker nodes, and each one of the plurality of request notification queues is associated with a respective one of the plurality of worker nodes;

store request data, response data and session data of the request in a memory, the memory being accessible to the connection manager and the worker node;

examine at a dispatcher whether the worker node assigned to handle the session is available to process the request, the examining including querying the session data to determine if the session is an existing session and retrieving the worker node assigned to handle the session;

send the request from the dispatcher to the assigned worker node if the assigned worker node is available, the worker node processing the request by retrieving the request notification from the request notification queue associated with the worker node and the request data and the session data from the memory; and

select another worker node to which to send the request according to a load balancing algorithm if the assigned worker node is not available.

19. The article of manufacture of claim 18 further comprising instructions that cause the processor to determine that the assigned worker node is available if there are no active requests for the session that the assigned worker node has yet to process.

20. The article of manufacture of claim 18, wherein the instructions that cause the processor to select another worker node to which to send the request comprises instructions that cause the processor to select on a weighted basis the other worker node from a list of worker nodes maintained in a data structure accessible to the dispatcher.

21. The article of manufacture of claim 20, wherein the instructions that cause the processor to select the other worker node is based on a lowest number of request notifications in a request notification queue associated with the other worker node as compared to a number of request notifications in a respective request notification queue associated with each of the worker nodes in the list.