Method and apparatus for making a cluster of computers appear as a single host on a network

• US 5,371,852 A
• Filed: 10/14/1992
• Issued: 12/06/1994
• Est. Priority Date: 10/14/1992
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Smith, Stephen E., Attanasio, Clement R.
• Primary Examiner(s): Shaw, Dale M.
• Assistant Examiner(s): Meky, Moustafa M.
• Application Number:
  US07/960,742
• Time in Patent Office: 783 Days
• Field of Search: 395/200, 395/500, 370/60, 370/92, 370/93, 370/94.1, 370/54, 370/85.13, 370/85.1, 370/85.6, 370/85.8, 370/60.1, 370/110.1, 370/85.11, 370/95.1, 364/284.4, 364/242.94, 364/284.3, 364/284
• US Class Current: 709/245
• CPC Class Codes: G06F 2209/505 Clust
• CPC Class Codes: G06F 9/5061 Partitioning or combining of resources
• CPC Class Codes: H04L 45/04 Interdomain routing, e.g. hierarchical routing
• CPC Class Codes: H04L 67/2895 Intermediate processing functionally located close to the data provider application, e.g. reverse proxies
• CPC Class Codes: H04L 67/563 Data redirection of data network streams

Abstract

The present invention provides a method and apparatus for enabling a cluster of computers to appear as a single computer to host computers outside the cluster. A host computer communicates only with a gateway to access destination nodes and processes within the cluster. The gateway has at least one message switch which processes incoming and outgoing port type messages crossing the cluster boundary. This processing comprises examining certain information on the message headers and then changing some of this header information either to route an incoming message to the proper computer node, port and process or to make an outgoing message appear as if originated at the gateway node. The message switch uses a table to match incoming messages to a particular routing function which can be run to perform the changes necessary to correctly route different kinds of messages.

Claims

1. A method for routing incoming messages across a boundary of a cluster of computer nodes, the cluster connected to one or more networks, comprising the steps of:

reading a software communication protocol number in a message header of the message to recognize an incoming message as a software communication protocol port type message, the message having a destination address of a gateway node within the cluster of computer nodes;

locating and reading a software communication protocol port number in the message header of the software communication protocol port type message;

matching both the software communication protocol port number and the software communication protocol number to an entry in a message switch memory, the matched software communication protocol port number entry being associated with a software communication protocol port specific function which selects a routing destination for the message from a plurality of possible destinations, the destination being a computer node in the cluster; and

routing the message to the computer node destination.

2. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 1, where the function modifies the destination address in the message header to that of the address of the destination node selected by the function.

3. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 2, where the function also modifies a destination port number in the message header to that of a port number on the destination node.

4. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 1, where there is no matched entry in the message switch memory and the destination node is computed by a routing algorithm.

5. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 4, where the algorithm determines the destination node address by using the destination port number in the message header.

6. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 1, where messages are sent to the selected destination node without changing the destination address field in the message header, and the destination node accepts any message having a destination address recognized by a cluster gateway message switch.

7. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 1, where the message which is to be routed to a failed node is routed to an operating node in the cluster.

8. A method for routing incoming messages across a boundary of a cluster of computer nodes, the cluster connected to one or more networks, comprising the steps of:

reading a software communication protocol number in an IP message header to recognize the incoming message as a software communication protocol port type message, the IP message header being on a message with a destination address of a gateway within the cluster of computer nodes;

locating and reading a software communication protocol port number in the message header of the software communication protocol port type message;

matching both the software communication protocol port number and the software communication protocol number to an entry in a message switch memory, the matched software communication protocol port number entry being associated with a port specific function which selects a routing destination for the message from a plurality of possible destinations, the destination being a computer node in the cluster; and

routing the message to the computer node destination.

9. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 8, where the function modifies the destination address in the message header to that of the address of the destination node selected by the function.

10. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 9, where the function also modifies the destination port number in the IP message header to that of a port number on the destination node.

11. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 8, where there is no matched entry in the message switch memory, and the destination is computed by a routing algorithm.

12. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 11, where the algorithm determines the destination node address by using the destination port number in the message header.

13. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 8, where the protocol number is a UDP identifier.

14. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 13, where the UDP protocol number is 17.

15. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 8, where there is more than one destination node in the cluster using a common port number.

16. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 15, where the common port number is the port number associated with NFS.

17. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 15, where the common port number is the port number associated with RLOGIN.

18. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 8, where the message is a MOUNT type request for a filehandle and part of the message is a data field which contains a filesystem name that was requested by the source host of the MOUNT request.

19. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 18, where the selected node returns the filehandle which includes a destination node identifier used by a later NFS request.

20. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 8, where the message is a NFS type request and the message data field contains a filehandle.

21. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 20, where the filehandle includes a destination node identifier.

22. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 8, where the protocol number is a TCP identifier.

23. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 22, where the TCP protocol number is 6.

24. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 22, where the function is a connection manager which routes the message to a destination computer node in the cluster based on the values of a source port number and a source IP address number.

25. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 24, where the destination computer node for new connections is determined by algorithm.

26. A method of routing incoming messages across a boundary of a cluster of computers, as in claim 25, where the algorithm determines the destination computer node in a way to balance a load among nodes in the cluster.

27. A method of routing incoming messages across a boundary of a cluster of computer, as in claim 26, where the algorithm is a round-robin.

28. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 25, where the algorithm ensures that, for each external source host, all TCP messages from that source host are routed to the same destination node.

29. A method of routing incoming messages across the boundary of a cluster of computers, as in claim 8, where messages are sent to the selected destination node without changing the destination IP address field in the message header, and the destination node accepts any message having a destination IP address recognized by a cluster gateway message switch.

30. A method of routing incoming messages, as in claim 12, where the software communication protocol number in the IP message header is a TCP protocol number and the software communication protocol port number is a TCP port number.

31. A method of routing an incoming TCP connection based message across a boundary of a cluster of computers, as in claim 30, where the message is a RLOGIN type.

32. A method of routing an incoming TCP connection based message across a boundary of a cluster of computers, as in claim 30, where the message is a REEXEC type.

33. A method of routing an incoming TCP connection based message across a boundary of a cluster of computers, as in claim 30, where the message is a RSH type.

34. A method of routing an incoming TCP connection based message across a boundary of a cluster of computers, as in claim 30, where the message is a TELNET type.

35. An apparatus for routing messages across a boundary of a cluster of computers on a network, comprising:

two or more computers connected together by an interconnect in a cluster, each computer being a node of the cluster;

one or more computer nodes in the cluster being a gateway node to the cluster, the gateway node being connected to the network at a cluster boundary and having a software communication protocol address on the network, the protocol address being the address of the gateway and the cluster;

a message switch, residing in the gateway node which matches both a message software communication protocol number and destination software communication protocol port number of the message coming into the cluster from the network to an entry in a table in the message switch memory; and

one or more routing functions, residing in the message switch table, one of the routing functions associated with the matched table entry which is used for directing messages crossing the cluster boundary into the cluster to a destination node within the cluster, the destination node being determined by the message switch using information on the message.