Load balancing of servers in a cluster

• US 7,389,510 B2
• Filed: 11/06/2003
• Issued: 06/17/2008
• Est. Priority Date: 11/06/2003
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: International Business Machines Corporation
• Inventors: Forrester, Jason D.
• Primary Examiner(s): An; Meng-Al T.
• Assistant Examiner(s): To; Jennifer N.
• Application Number:
  US10/703,900
• Publication Number:
  US 20050102676A1
• Time in Patent Office: 1,685 Days
• Field of Search: 718100-105, 709223-226, 709/249
• US Class Current: 718/105
• CPC Class Codes: G06F 2209/505 Clust
• CPC Class Codes: G06F 9/505 considering the load

Abstract

A system, method and program product for operating a cluster of servers. A load balancer, separate from the servers, selects servers to handle respective work requests assigned to the cluster based on a load balancing algorithm, and sends the work requests to the selected servers. The load balancer receives a request from another server not currently a member of the cluster. The request requests that the other server become a member of the cluster to handle work requests. In response to the request, the load balancer joins the other server as a member of the cluster to handle some work requests subsequently assigned to the cluster. The other server specifies to the load balancer a load threshold for the other server. The other server specifies to the load balancer a time-out to be used for the other server. The time-out indicates a time allotted for the other server to send a hello message since sending a previous hello message.

Claims

1. A method for managing a cluster of servers, said method comprising the steps of:

a load balancer selecting servers to handle respective work requests assigned to said cluster based on a load balancing algorithm, and sending said work requests to the selected servers;

said load balancer receiving a membership request from another server not currently a member of said cluster, said membership request requesting that said other server become a member of said cluster to handle work requests and specifying a threshold level of work requests that can be handled by said other server, said load balancer not resident in said other server, wherein said threshold level of work requests is a number of message packets over a unit of time; and

in response to said membership request from said other server, said load balancer joining said other server as a member of said cluster to handle some work requests subsequently assigned to said cluster based on said load balancing algorithm and said threshold level of work requests for said other server.

2. A method as set forth in claim 1 wherein said other server specifies to said load balancer a preferred load balancing algorithm to be used for assigning work requests to servers in said cluster.

3. A method as set forth in claim 2 wherein there are other servers already in said cluster and said load balancer determines that said load balancer is currently using a different load balancing algorithm for said other servers, and in response, said load balancer rejects said preferred load balancing algorithm.

4. A method as set forth in claim 1 wherein said load balancer subsequently sends work requests to said other server based on said load balancing algorithrm and said threshold level of work requests for said other server.

5. A method as set forth in claim 1 further comprising the steps of:

said load balancer subsequently receving another request from a third server not currently a member of said cluster, said other request reguesting that said third server become a member of said cluster to handle work requests and specifving another threshold level of work requests that can be handled by said third servers; and

wherein said load balancer is using a lower threshold level of work requests for all other servers in said cluster or has a lower default threshold level of work request for servers in said cluster, and in response, said load balancer rejecting said other load threshold level of work requests specified by said third server, and further comprising the steps of;

in response to said other request, said load balancer joining said third server as a member of said cluster to handle some work requests subsequently assigned to said cluster based on said load balancing algorithrn and said lower threshold level of work requests for all other servers in said cluster or said lower default threshold level of work requests for servers in said cluster.

6. A method as set forth in claim 1 wherein said other server specifies to said load balancer a time-out to be used for said other server, said time-out indicating a time allotted for said other server to send a hello message since sending a previous hello message;

and further comprising the steps of;

said load balancer adopting said time-out for said other server and waiting said time for said other server to send a hello message since sending a previous hello message before removng said other server from said cluster.

7. A method as set forth in claim 1 wherein:

said other server specifies to said load balancer a time-out to be used for said other server, said time-out indicating a time allotted for said other server to send a hello message since sendng a previous hello message; and

said load balancer is using a shorter time-out for all servers in said cluster or has a shorter default time-out for servers in said cluster, and in response, said load balancer rejecting said time-out specified by said other server.

8. A method as set forth in claim 1 wherein said other server specifies to said load balancer a hello interval to be used for said other server, said hello interval indicating that said other server is still available.

9. A method as set forth in claim 8 wherein said load balancer is using a shorter hello interval for all servers in said cluster or has a shorter default hello interval for servers in said cluster, and in response, said load balancer rejecting said hello interval specified by said other server.

10. A method as set forth in claim 1 wherein said load balancer executes in a computing device that is not one of said servers in said cluster which handles said work requests.

11. A method as set forth in claim 1 further comprising the steps of:

said load balancer checking for a periodic hello message from each server in said cluster, and wherein said periodic hello message is not received by said load balancer for one of said servers in said cluster within a predetermined time-out period, and in response, said load balancer removing said one server from said cluster.

12. A method as set forth in claim 1 further comprising the steps of:

said load balancer checking for a periodic hello message from each server in said cluster, said periodic hello message including authentication information from said each server, and wherein said periodic hello message is received from one of said servers but said authentication information for said one server is invalid, and in response, said load balancer removing said one server from said cluster.

13. A method as set forth in claim 1 further comprising the subsequent steps of:

said load balancer receiving a request from said other server to be removed from said cluster, and in response, said load balancer removing said other sever from said cluster.

14. A method as set forth in claim 1 further comprising the steps of:

in response to a workload of said cluster of servers or a workload of a server in said cluster reaching a predetermined threshold, said load balancer sending an invitation to one or more additional servers to join said cluster; and

said load balancer responding to acceptance of said invitation from one or more additional servers by joining to said cluster the one or more additional servers that accepted said invitation.

15. A method as set forth in claim 14 further comprising the steps of:

in response to said workload of said cluster or said workload of said servers decreasing below a predetermined threshold, said load balancer sending a request to at least one server in said cluster to be removed from said cluster.

16. A method as set forth in claim 1 wherein said load threshold for work requests for said other server is different than a load threshold for work requests for servers that were members of said cluster when said other server became a member of said cluster.

17. A computer program product for managing a cluster of servers, said computer program product comprising:

a computer readable media;

first program instructions to select servers to handle respective work requests assigned to said cluster based on a load balancing algorithm, and initiate sending of said work requests to the selected servers;

second program instructions to receive a membership request from another server not currently a member of said cluster, said membership request requesting that said other server become a member of said cluster to handle work requests and specifying a threshold level of work requests that can be handled by said other server, wherin said threshold level of work requests is a number of message packets over a unit of time; and

third program instructions, responsive to said membership request from said other server, to join said other server as a member of said cluster to handle some work requests subsequently assigned to said cluster based on said load balancing algorithm and said threshold level of work requests for said other server; and

wherein

said first, second and third program instructions are executed on a computing device other than said other server; and

said first, second and third program instructions are stored on said media.

18. A computer program product as set forth in claim 17 wherein said load threshold for work requests for said other server is different than a load threshold for work requests for servers that were members of said cluster when said other server became a member of said cluster.

19. A computer for managing a cluster of servers, said computer comprising:

a central processing unit;

first program instructions to select servers to handle respective work requests assigned to said cluster based on a load balancing algorithm, and initiate sending of said work requests to the selected servers;

second program instructions to receive a membership request from another server not currently a member of said cluster, said membership request requesting that said other server become a member of said cluster to handle work requests and specifying a threshold level of work requests that can be handled by said other server, wherein said threshold level of work requests is a number of message packets over a unit of time; and

third program instructions, responsive to said membership request from said other server, to join said other server as a member of said cluster to handle some work requests subsequently assigned to said cluster based on said load balancing algorithm and said threshold level of work requests for said other server; and

wherein

said first, second and third program instructions are stored in said computer for execution by said central processing unit.

20. A computer as set forth in claim 19 wherein said load threshold for work requests for said other server is different than a load threshold for work requests for servers that were members of said cluster when said other server became a member of said cluster.