System partitioning for massively parallel processors

• US 5,799,149 A
• Filed: 06/17/1996
• Issued: 08/25/1998
• Est. Priority Date: 06/17/1996
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Brenner, Larry Bert, Rothaupt, Krystal Kay, Briskey, Kenneth Charles
• Primary Examiner(s): Geckil, Mehmet B.
• Application Number:
  US08/664,900
• Time in Patent Office: 799 Days
• Field of Search: 395/200.31, 395/200.45, 395/200.74, 395/200.75, 707/103, 707/10, 711/153, 711/173
• US Class Current: 709/201
• CPC Class Codes: G06F 9/5077 Logical partitioning of resources; Management or configuration of virtualized resources specific details on emulation or internal functioning of virtual machines G06F9/455

Abstract

In accordance with a preferred embodiment of the present invention, a method and apparatus is disclosed for partitioning a massively distributed parallel processor into at least two isolated sub-environments. The parallel processing environment is comprised of a plurality of nodes connected to one another via a network in such a way as to form a single system unit initially. A central control element is provided for controlling and maintaining all communications and status information between nodes. The nodes and the central control element are in processing communication with one another via the network. All system data is deposited in a system data repository located in the central control element. Once the environment is partitioned, the system data stored in this repository is organized into system or partitioned classes, with the partitioned classes only being accessible selectively to nodes residing in one environment or another.

Claims

1. In a distributed parallel processing environment having a plurality of nodes connected to one another via a network in such way as to form a single system unit, a method of logically partitioning said processing environment into at least two isolated sub-environments, said method comprising the steps of:

controlling all communications through a central control element connected to said nodes via said network;

storing all system data in a system data repository located in said central control element;

identifying each node as either belonging to a first sub-environment or a second sub-environment; and

organizing data deposited in said system data repository into system object classes providing data globally accessible to all nodes, and partitioned object classes providing data that is exclusive to each sub-environment.

2. The method of claim 1, further comprising the step of said central control element providing a client, only data either classified in said system object classes or data pertaining to said client'"'"'s respective sub-environment, said client being defined as any node on either of said sub-environments querying said central control element.

3. The method of claim 2, wherein said nodes in said first sub-environment can only access data provided in said partitioned object classes belonging to said first sub-environment, and said nodes in said sub-environment can only access data provided in said partitioned object classes belonging to said second sub-environment.

4. The method of claim 3, wherein data relating to configuration of said processing environment is classified as system object class.

5. The method of claim 1, wherein said processing environment is a UNIX based operating environment.

6. The method of claim 1, wherein said network is an ethernet using a TCP/IP protocol.

7. The method of claim 1, wherein a plurality of nodes are connected to each other via a switch network as well as via said network, said switch network being also connected to said other network but only in processing communication with said central control element via said network.

8. The method of claim 7, wherein said switch network comprises of at least two switches in direct communication with one another and with said nodes, in a manner as to provide an additional second processing network to said nodes.

9. The method of claim 8, wherein said switches are high performance switches.

10. The method of claim 8, further comprising the step of partitioning said switches, so that each switch belongs exclusively to only one sub-environment.

11. The method of claim 10, further comprising the step of creating a switch topology file for each system sub-environment.

12. The method of claim 11, further comprising the step of creating a switch initialization file for each system sub-environment.

13. The method of claim 12, wherein said switch initialization files and said switch topology files for each said partition is included as part of data exclusively available only to said sub-system environment.

14. The method of claim 13, wherein all switch faults and all switch message traffic is also provided for separately for each sub-environment.

15. The method of claim 10, wherein at least one node can be deleted from said system unit and said respective sub-environment further comprising the step of updating said system data in said system repository to reflect deletion of said new node.

16. The method of claim 10, wherein at least one switch can be deleted from said system unit and said respective sub-environment further comprising the step of updating said system data in said system repository to reflect deletion of said new switch.

17. The method of claim 10, wherein said sub-environments can be redefined so that nodes originally belonging to a first sub-environment can be reassigned to said second sub-environment and nodes originally belonging to said second sub-environment can be reassigned to said first sub-environment, further comprising the step of updating said system data in said system repository to reflect these changes.

18. The method of claim 10, wherein said sub-environments can be redefined so that switches originally belonging to a first sub-environment can be reassigned to said second sub-environment and switches originally belonging to said second sub-environment can be reassigned to said first sub-environment, further comprising the step of updating said system data in said system repository to reflect these changes.

19. The method of claim 1, further comprising the step of grouping a plurality of nodes in a frame, and providing each frame at least one individual switch, in a manner so that said processing environment comprises at least two frames and two switches.

20. The method of claim 19, wherein said switches are high performance switches.

21. The method of claim 1, wherein said network is an ethernet using a TCP/IP protocol.

22. The method of claim 1, wherein a resource can be added to said system unit and respectively to any sub-environment further comprising the step of updating said system data in said system repository after said addition to reflect addition of said new resource.

23. The method of claim 1, wherein at least one node can be deleted from said system unit and said respective sub-environment further comprising the step of updating said system data in said system repository to reflect deletion of said new node.

24. The method of claim 1, wherein said sub-environments can be redefined so that nodes originally belonging to a first sub-environment can be reassigned to said second sub-environment and nodes originally belonging to said second sub-environment can be reassigned to said first sub-environment, further comprising the step of updating said system data in said system repository to reflect these changes.

25. A massively parallel processing environment partitioned into sub-environments and having a plurality of nodes connected via a network in such a way as to initially create a single system unit, said processing environment comprising:

a central control element for controlling all communications between said nodes and for maintaining configuration and status information, said central control element being in processing communications with said nodes via said network;

at least two partitioned sub-environments each having exclusive ownership of a subset of nodes in a manner that no two nodes in said system unit environment belong to said same sub-environment;

a system data repository located in said central control element for storing system data, said data in said repository organized into a plurality of object classes; and

said object classes being divided into system and partitioned classes, said system classes having data targeted to be globally available to all nodes on both sub-environments and said data in said partitioned classes being exclusively accessible only to nodes of one sub-environment or another.

26. The processing environment of claim 25, wherein said system unit operate in a UNIX based environment.

27. The processing environment of claim 25, wherein said network is an ethernet using a TCP/IP protocol.

28. The processing environment of claim 25, wherein a plurality of nodes are connected to each other via a switch network as well as via said network, said switch network being also connected to said network, but only in processing communication with said central control element via said other network.

29. The processing environment of claim 28, wherein said switch network comprises a plurality of switches in direct communication with one another as to provide a second processing network to said nodes.

30. The processing environment of claim 29, wherein said switches are high performance switches.

31. The processing environment of claim 30, wherein said other network is an ethernet using a TCP/IP protocol.

32. The processing environment of claim 31, further comprising partitioning said switches, so that each switch belongs exclusively to only one sub-environment.

33. The processing environment of claim 25, further comprising grouping a plurality of nodes in a frame, said frame having at least one individual switch, in a manner so that said system environment comprises at least two frames and two switches.