Method and computer program product for controlling distributed-memory multiprocessor system

• US 6,092,098 A
• Filed: 07/24/1998
• Issued: 07/18/2000
• Est. Priority Date: 07/25/1997
• Status: Expired due to Term

• Current Assignee: Rakuten, Inc. (Rakuten Group, Inc.)
• Original Assignee: NEC Corporation
• Inventors: Araki, Hiroyuki, Tatsukawa, Kosuke
• Primary Examiner(s): Harrell, Robert B.
• Application Number:
  US09/121,826
• Time in Patent Office: 725 Days
• Field of Search: 364/DIG. 1, 364/DIG. 2, 709/100, 709/102, 709/104, 709/105, 709/106-201, 708/446, 712/28, 717/6
• US Class Current: 709/201
• CPC Class Codes: G06F 9/461 Saving or restoring of program or task context
• CPC Class Codes: G06F 9/544 Buffers; Shared memory; Pipes

Abstract

Disclosed is a method for controlling a distributed-memory multiprocessor system having a plurality of nodes mutually connected by a network, each of the nodes having a CPU and a memory, and a plurality of user tasks which belong to the same application and are executed in different nodes, which method comprises the steps of: (a) assigning a transmission resource of a network interface in the node to a user task that will be executed next and starting the execution of the user task when a context switching operation is performed; (b) reading data written by the user task from the transmission resource, adding to the data a communication identifier that is assigned to an application to which the user task belongs and that identifies a data reception area, and transmitting the resultant data to a reception side node through the network, in the network interface of each transmission side node; and (c) selecting a data reception area for storing data received from another node from a plurality of data receiving areas on the basis of the communication identifier added to the received data and storing the received data in the selected data reception area, in the network interface of each reception side node.

Claims

1. A method for controlling a distributed-memory multiprocessor system having a plurality of nodes mutually connected by a network, each of said nodes having a CPU and a memory, and a plurality of user tasks which belong to the same application and are executed in different nodes, which method comprises the steps of:

(a) assigning a transmission resource of a network interface in said node to a user task that will be executed next and starting the execution of said user task when a context switching operation is performed;

(b) reading data written by said user task from said transmission resource, adding to said data a communication identifier that is assigned to an application to which said user task belongs and that identifies a data reception area, and transmitting the resultant data to a reception side node through said network, in said network interface of each transmission side node; and

(c) selecting a data reception area for storing data received from another node from a plurality of data receiving areas on the basis of said communication identifier added to said received data and storing said received data in said selected data reception area, in said network interface of each reception side node.

2. The method as set forth in claim 1,

wherein the step (a) comprises the step of

(a1) assigning a dummy transmission resource to said user task which will be executed next so that said user task can be executed when said network interface does not have a transmission resource that can be assigned to said user task in the context switching operation.

3. The method as set forth in claim 2,

wherein the step (a1) ) is followed by the step of:

(a2) re-determining whether or not said network interface has a transmission resource that can be assigned to said user task when the user task accesses said dummy transmission resource, assigning said transmission resource instead of said dummy transmission resource to said user task when said network interface has the transmission resource that can be assigned to said user task, and continuously executing said user task.

4. The method as set forth in claim 1,

wherein the step (a) comprises the step of:

(a1) storing the relation between said communication identifier for said user task which will be executed next and a pointer which points to said data reception area for said user task to a cache of said local network interface on said reception side node; and

wherein the step (c) comprises the step of

(c1) retrieving said data reception area for storing said received data from said cache using said communication identifier added to said received data as a key, and selecting said data reception area to which said received data is written.

5. The method as set forth in claim 4,

wherein said network interface generates interrupt which causes said OS to store said received data when there is not a cache having the communication identifier identical with said identifier added to said received data.

6. The method as set forth in claim 3,

wherein the step (a2) is performed only when the attribute of said user task is other than the attribute of Local which represents that said user task does not execute communication using said network interface.

7. The method as set forth in claim 1,

wherein the step (a) is performed by assigning a dummy transmission resource to said user task which will be executed next and starting execution of said user task when said user task has the attribute of Local which represents that said user task does not execute communication using said network interface.

8. The method as set forth in claim 1,

wherein said data reception area is a page in a virtual address space and a page table stores the address of said page as a data reception area;

and

wherein the step (c) is performed by retrieving a page as a data receiving area for storing received data from said page table using said communication identifier added to said received data as a key, and selecting said data reception area, in said network interface of said each reception side node.

9. The method as set forth in claim 1, further comprising the step of:

(d) causing a CID manager for managing said communication identifiers to assign a communication identifier to an application before said application performs a communicating process and;

(e) causing said CID manager to collect said assigned communication identifier from said application after completion of said communicating process so that said collected communication identifier can be reused.

10. The method as set forth in claim 1,

wherein the system has a scheduler, said scheduler having a sequencer and stubs, said sequencer managing mapping of said applications and said user tasks that belong thereto and determining a schedule for a group of tasks executed across said nodes, each of said stubs being disposed in each of said nodes and performing said context switching operation for said tasks in each of said node corresponding to the command received from said sequencer, said sequencer sending said schedule to said stubs asynchronously with said context switching operation.

20. The computer program product as set forth in claim 1,

wherein the system has a scheduler, said scheduler having a sequencer and stubs, said sequencer managing mapping of said applications and said user tasks that belong thereto and determining a schedule for a group of tasks executed across said nodes, each of said stubs being disposed in each of said nodes and performing said context switching operation for said tasks in each of said node corresponding to the command received from said sequencer, said sequencer sending said schedule to said stubs asynchronously with said context switching operation.

11. A computer program product which causes a distributed-memory multiprocessor system having a plurality of nodes mutually connected by a network, each of said nodes having a CPU and a memory, and a plurality of user tasks which belong to the same application and are executed in different nodes, to execute a method which comprises the steps of:

(a) assigning a transmission resource of a network interface in said node to a user task that will be executed next and starting the execution of said user task when a context switching operation is performed;

(b) reading data written by said user task from said transmission resource, adding to said data a communication identifier that is assigned to an application to which said user task belongs and that identifies a data reception area, and transmitting the resultant data to a reception side node through said network, in said network interface of each transmission side node; and

(c) selecting a data reception area for storing data received from another node from a plurality of data receiving areas on the basis of said communication identifier added to said received data and storing said received data in said selected data reception area, in said network interface of each reception side node.

12. The computer program product as set forth in claim 11,

wherein the step (a) comprises the step of

(a1) assigning a dummy transmission resource to said user task which will be executed next so that said user task can be executed when said network interface does not have a transmission resource that can be assigned to said user task in the context switching operation.

13. The computer program product as set forth in claim 12,

wherein the step (a1) is followed by the step of:

(a2) re-determining whether or not said network interface has a transmission resource that can be assigned to said user task when the user task accesses said dummy transmission resource, assigning said transmission resource instead of said dummy transmission resource to said user task when said network interface has the transmission resource that can be assigned to said user task, and continuously executing said user task.

14. The computer program product as set forth in claim 13,

wherein the step (a2) is performed only when the attribute of said user task is other than the attribute of Local which represents that said user task does not execute communication using said network interface.

15. The computer program product as set forth in claim 11,

wherein the step (a) comprises the step of:

(a1) storing the relation between said communication identifier for said user task which will be executed next and a pointer that points to said data reception area for said user task to a cache of said local network interface on said reception side node; and

wherein the step (c) comprises the step of

(c1) retrieving said data reception area for storing said received data from said cache using said communication identifier added to said received data as a key, and selecting said data reception area to which said received data is written.

16. The computer program product as set forth in claim 15,

wherein said network interface generates interrupt which causes said OS to store said received data when there is not a cache having the communication identifier identical with said identifier added to said received data.

17. The computer program product as set forth in claim 11,

wherein the step (a) is performed by assigning a dummy transmission resource to said user task which will be executed next and starting execution of said user task when said user task has the attribute of Local which represents that said user task does not execute communication using said network interface.

18. The computer program product as set forth in claim 11,

wherein said data reception area is a page in a virtual address space and a page table stores the address of said page as a data reception area;

and

wherein the step (c) is performed by retrieving a page as a data receiving area for storing received data from said page table using said communication identifier added to said received data as a key, and selecting said data reception area, in said network interface of said each reception side node.

19. The computer program product as set forth in claim 11,

wherein said method further comprising the step of:

(d) causing a CID manager for managing said communication identifiers to assign a communication identifier to an application before said application performs a communicating process and to collect said assigned communication identifier from said application so that said collected communication identifier can be reused.