Operating system for use with computer networks incorporating two or more data processors linked together for parallel processing and incorporating improved dynamic load-sharing techniques

• US 6,078,945 A
• Filed: 07/28/1997
• Issued: 06/20/2000
• Est. Priority Date: 06/21/1995
• Status: Expired due to Fees

• Current Assignee: Tao Group Limited (Tao Media Ltd.)
• Original Assignee: Tao Group Limited (Tao Media Ltd.)
• Inventors: Hinsley, Christopher Andrew
• Primary Examiner(s): Banankhah, Majid A.
• Application Number:
  US08/901,564
• Time in Patent Office: 1,058 Days
• Field of Search: 709/100, 709/102, 709/105, 709/106, 709/108
• US Class Current: 718/105
• CPC Class Codes: G06F 9/5088 involving task migration

Abstract

A data processing system incorporates two or more processors linked together in a network which are capable of parallel processing. The data processing system employs a load-sharing technique where any given processor, upon allocation of a process, decides based on the loading of its immediate neighbors in the network, whether it or one of its neighboring processors is better able at that time to execute that process. In accordance with that decision, the given processor either executes the process itself or hands the process over to one of its neighbors if that neighbor is better able at that time to perform that process.

Claims

1. A data processing system comprising a plurality of data processors interconnected as nodes in a random network of data processors and arranged to perform processes in parallel, each processor being directly connected in the network to a number of neighboring processors located at adjacent nodes of the network, wherein each of the processors in the network is programmed to decide, after being allocated a process, whether that processor or one of the number of neighboring processors, can at that time more optimally perform said process, and in accordance with that decision, either to reallocate said process to the one neighboring processor if the one neighboring processor can more optimally perform said process, or to perform the process itself if that processor decides that none of the neighboring processors can more optimally perform said process.

2. A system as claimed in claim 1, wherein upon being allocated a process:

each processor is programmed to compare a rating of that processor with a corresponding rating from each of the neighboring processors, those ratings representing each individual processor'"'"'s ability at that time to perform said process;

to determine which of the processors has the highest rating; and

to reallocate said process to the neighboring processor having the highest rating or to perform the process itself if that processor has a rating equal to or higher than that of any of the neighboring processors.

3. A system as claimed in claim 2, wherein the rating of each processor is a function of the number of processes currently operating on that processor.

4. A system as claimed in claim 2, wherein the rating of each processor is represented by the effective operations-per-second rating of that processor divided by the number of processes running at that processor and multiplied by a function of the off-chip communication speed available to that processor.

5. A system as claimed in claim 2, wherein the rating of each processor is a function of the available local memory space on that processor.

6. A system as claimed in claim 2, wherein, upon allocating the process to a neighboring processor, the allocating processor is programmed to provide the neighboring processor with the number of bytes required, the name of the process, and an address at which the process is currently stored.

7. A system as claimed in claim 2, wherein information relating to the ability of any given processor in the network to perform processes at any given time is regularly added to other messages passing between that processor and that processor'"'"'s immediate neighbors.

8. A system as claimed in claim 2, wherein information relating to the ability of any given processor in the network to perform a process at any given time is passed to that processor'"'"'s immediate neighbors at a time when no other information is being passed between those processors.

17. A data processing system as claimed in claim 1, which is programmed to repeat the allocation process at least once, so that the one neighboring processor, upon being allocated said process, is programmed to decide on the basis of information currently available to that one neighboring processor concerning that one neighboring processor'"'"'s own ability to perform the process and the ability of its neighbors to perform the process, whether that one neighboring processor or one of its neighboring processors is even more optimally suited at that time to perform the process, and in accordance with that decision, either to transfer the process yet again to another neighboring processor if that other neighboring processor is even more optimally suited at that time to perform said process, or to perform the process itself if that one neighboring processor decides that it can still most optimally perform the process.

19. A system as claimed in claim 1, wherein the number of neighboring processors to which each processor in the network is directly connected is less than the total number of processors in the network.

9. A method of processing data in a data processing system including a plurality of electronic data processors connected as nodes in a random network of data processors and capable of performing processes in parallel, each processor being directly connected in the network to a number of neighboring processors located at the adjacent nodes of the network, wherein upon allocation of a process to any given processor in the network, the given processor performs the following steps:

(a) decides on the basis of information currently available to that processor whether that processor or one of the number of neighboring processors is best suited at that time to perform the process; and

(b) in accordance with the decision in step (a) either performs the process itself or transfers the process to the one neighboring processor.

10. A method as claimed in claim 9, wherein upon being allocated a process:

each processor compares a rating of that processor with a corresponding rating from each of the neighboring processors, those ratings representing each individual processor'"'"'s ability at that time to perform said process;

determines which of the processors has the highest rating; and

reallocates said process to the neighboring processor having the highest rating or performs the process itself if that processor has a rating equal to or higher than that of any of the neighboring processors.

11. A method as claimed in claim 10, wherein the rating of each processor is a function of the number of processes currently operating on that processor.

12. A method as claimed in claim 10, wherein the rating of each processor is represented by the effective operations-per-second rating of that processor divided by the number of processes running at that processor and multiplied by a function of the off-chip communication speed available to that processor.

13. A method as claimed in claim 10, wherein the rating of each processor is a function of the available local memory space on that processor.

14. A method as claimed in claim 10, wherein, upon allocating the process to a neighboring processor, the allocating processor provides the neighboring processor with the number of bytes required, the name of the process, and an address at which the process is currently stored.

15. A method as claimed in claim 10, wherein information relating to the ability of any given processor in the network to perform processes at any given time is regularly added to other messages passing between that processor and that processor'"'"'s immediate neighbors.

16. A method as claimed in claim 10, wherein information relating to the ability of any given processor in the network to perform a process at any given time is passed to that processor'"'"'s immediate neighbors at a time when no other information is being passed between those processors.

18. A method according to claim 9, wherein following the transfer of the process by the given data processor in step b) to the one neighboring processor, the said one neighboring processor decides, on the basis of information currently available to that processor concerning that one neighboring processor'"'"'s own ability to perform the process and the ability of its neighbors to perform the process, whether that one neighboring processor or one of its neighbors is even more optimally suited at that time to perform the process and, in accordance with that decision, either transfers the process yet again to that one of its neighbors which is even more optimally suited at that time to perform the process, or performs the process itself if that one neighboring processor is still the one most optimally suited to perform the process.

20. A system as claimed in claim 11, wherein the number of neighboring processors to which each processor in the network is directly connected is less than the total number of processors in the network.

21. A method according to claim 9, wherein the number of neighboring processors to which each processor in the network is directly connected is less than the total number of processors in the network.

22. A method as claimed in claim 18, wherein the number of neighboring processors to which each processor in the network is directly connected is less than the total number of processors in the network.