Software architecture for control of highly parallel computer systems

• US 5,692,193 A
• Filed: 03/31/1994
• Issued: 11/25/1997
• Est. Priority Date: 03/31/1994
• Status: Expired due to Term

• Current Assignee: NEC Corporation
• Original Assignee: NEC Research Institute, Inc. (NEC Corporation)
• Inventors: Jagannathan, Suresh, Philbin, James F.
• Primary Examiner(s): Kriess, Kevin A.
• Application Number:
  US08/221,026
• Time in Patent Office: 1,335 Days
• Field of Search: 395/700, 395/650, 395/406, 395/676, 395/683
• US Class Current: 718/106
• CPC Class Codes: G06F 8/451 Code distribution considering CPU load at run-time G06F9/505; load rebalancing G06F9/5083

Abstract

A computer software architecture for controlling a highly parallel computer system comprises several layers of abstraction. The first layer is an abstract physical machine which contains a set of abstract physical processors. This layer may be considered as a microkernel. The next layer includes virtual machines and virtual processors. A virtual machine comprises a virtual address space and a set of virtual processors that are connected in a virtual topology. Virtual machines are mapped onto abstract physical machines with each virtual processor mapped onto an abstract physical processor. The third layer of abstraction defines threads. Threads are lightweight processes that run on virtual processors. In a preferred embodiment the abstract physical machines, abstract physical processors, virtual machines, virtual processors, thread groups, and threads are all first class objects.

Claims

1. A software architecture for controlling a highly parallel computer system comprising:

abstract physical machines comprising abstract physical processors forming a microkernel;

virtual machines associated with respective abstract physical processors, said virtual machines comprising virtual processors; and

thread groups comprising threads which run on said virtual processors,

where said virtual processors and said threads are first class objects.

2. A software architecture as set forth in claim 1, where said virtual processors are connected in a virtual topology.

3. A software architecture as set forth in claim 1, where said microckernel policy manager is user customizable.

4. A software architecture as set forth in claim 1, where said virtual processors contain which thread policy managers are user customizable.

5. A software architecture as set forth in claim 1, where said threads, said virtual processors and said abstract physical processors are operatively associated for constructing a virtual topology.

6. A software architecture as set forth in claim 5, where said virtual topology is user customizable.

7. A software architecture as set forth in claim 1, where said threads are separable from their respective execution contexts for allowing delayed allocation of execution contexts.

8. A software architecture as set forth in claim 1, further comprising polymorphic ports.

9. A architecture as set forth in claim 8, where said is a class object.

10. A software architecture as set forth in claim 8, where said threads send messages containing general data and complex data.

11. A software architecture as set forth in claim 1, where said threads garbage collect their respective local stacks and heaps independent of other threads.

12. A software architecture as set forth in claim 1, where thread groups collect their respective shared heaps independently of unrelated thread groups.

13. A software architecture as set forth in claim 1, where said virtual processors are multiplexed on said abstract physical processors.

14. A software architecture as set forth in claim 1, where said virtual processors said virtual machines and said threads reside in a persistent memory.

15. A software architecture as set forth in claim 1, where said abstract physical processors are first class objects.

16. A software architecture as set forth in claim 15, where said virtual machines are first class objects.

17. A software architecture as set forth in claim 16, where said abstract physical machines and said thread groups are first class object.

18. A computer system comprising:

a plurality of customizable abstract physical processors connected in a customizable physical topology each containing a customizable virtual processor controller and a customizable virtual processor policy manager;

a plurality of virtual machines each comprising a virtual address space whose topology is user specifiable and further comprising virtual processors which execute responsive to said virtual processor controller and said virtual processor policy manager and which contain a thread controller and a thread policy manager, said virtual processors being connected in a dynamic virtual topology and each virtual processor being mapped onto a respective abstract physical processor which mapping may be dynamically altered without modification to the implementation; and

a plurality of threads for running on said virtual processors responsive to said thread controllers and thread policy managers.

19. A computer system as set forth in claim 18, where said virtual processors are multiplexed on said abstract physical processors.

20. A computer system as set forth in claim 18, further comprising a persistent memory in which objects, including said threads, said virtual processors and said virtual machines, reside.

21. A computer system as set forth in claim 18, where the correspondence between threads and virtual processors and between virtual processors and physical processors and between virtual machines and physical machines are dynamically alterable.