Virtual processor method and apparatus for enhancing parallelism and availability in computer systems

• US 5,640,584 A
• Filed: 12/12/1994
• Issued: 06/17/1997
• Est. Priority Date: 12/12/1994
• Status: Expired due to Term

• Current Assignee: Teradata US, Inc. (Teradata Corporation)
• Original Assignee: NCR Corporation
• Inventors: Catozzi, John R., Kandasamy, David R., Heying, Douglas W.
• Primary Examiner(s): Pan, Daniel H.
• Application Number:
  US08/353,590
• Time in Patent Office: 918 Days
• Field of Search: 395/700, 395/650, 395/725, 395/480, 395/200.01, 395/500, 395/800, 395/200.18, 395/737, 395/375, 395/506, 395/508, 395/672, 395/200.15, 395/733, 395/200.16 , 395/676, 395/856, 395/674, 395/675, 395/735, 364/DIG. 1, 364/DIG. 2
• US Class Current: 712/30
• 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
• CPC Class Codes: G06F 9/546 Message passing systems or structures, e.g. queues

Abstract

A virtual processor method and apparatus for parallel computer systems that increases the level of parallelism to include multiple threads per node. If a processor node has a plurality of storage devices attached, a single thread can be allocated to each device. Similarly, if a processor node has multiple CPUs, each individual thread can utilize a different CPU. Thus, a task could potentially occupy all available hardware in the system. The result is increased system utilization and availability.

Claims

1. A computer system, comprising:

(a) a plurality of processor nodes, each of the processor nodes having a plurality of data storage devices connected thereto;

(b) an interconnect network, coupled to the processor nodes, for transmitting messages between processor nodes;

(c) each of the processor nodes comprising a plurality of vproc means for performing a plurality of threads simultaneously, wherein each vproc means independently performs one of the threads, the threads being streams of instructions executed on behalf of a task, wherein each of the threads access data on a different data storage device simultaneously with the other threads; and

(d) each of the processor nodes comprising operating system means for assigning the vproc means to processor nodes at system startup, further comprising means for creating and storing a translation table identifying locations of vproc means and processor nodes on the interconnect network at the system startup.

2. A computer system as recited in claim 1, wherein each processor node consists of one central processing unit (CPU).

3. A computer system as recited in claim 1, wherein:

each processor node has a plurality of central processing units (CPUs); and

said operating system means assigns the vproc means to processor nodes independent of the number of central processing units (CPUs) contained within each processing node.

4. A computer system as recited in claim 1, further comprising an interconnect network driver for transmitting messages between vproc means, wherein each of the vproc means includes one or more mailboxes, such that when a message is sent to a mailbox at a particular vproc means, the interconnect network driver looks up a vproc means/processor node translation in the translation table and sends the message to a designated processor node and the designated processor node routes it locally to an appropriate mailbox in the particular vproc means.

5. A computer system as recited in claim 1, further comprising a clustering of processor nodes connected to a common set of data storage devices, wherein the processor nodes are cross-connected to the common set of data storage devices.

6. A computer system as recited in claim 5, wherein each vproc means is connected to at least one virtual disk comprising one or more memory regions of the data storage devices that are grouped together, wherein the virtual disk is accessible by any processor node within a clique, thereby allowing the vproc means to be started and executed on any processor node within a clique.

7. A computer system, comprising:

(a) a plurality of processor nodes, each of the processor nodes having a plurality of data storage devices connected thereto;

(b) an interconnect network, coupled to the processor nodes, for transmitting messages between processor nodes;

(c) each of the processor nodes comprising a plurality of vproc means for performing a plurality of threads simultaneously, wherein each vproc means independently executes one of the threads, the threads being streams of instructions executed on behalf of a task, wherein each of the threads access data on a different data storage device simultaneously with the other threads; and

(d) each of the processor nodes comprising an operating system means for assigning the vproc means to processor nodes at system startup, for creating and storing a translation table identifying locations of vproc means and processor nodes on the interconnect network at startup time, and for providing updates to the translation table when movement of the vproc means between processor nodes occurs during runtime.

8. A computer system as recited in claim 7, wherein each processor node consists of one central processing unit (CPU).

9. A computer system as recited in claim 7, wherein:

each processor node has a plurality of central processing units (CPUs); and

said operating system means assigns the vproc means to processor nodes independent of the number of central processing units (CPUs) contained within each processing node.

10. A computer system as recited in claim 7, further comprising an interconnect network driver for transmitting messages between vproc means, wherein each vproc means includes one or more mailboxes, such that when a message is sent to a mailbox at a particular vproc means, the interconnect network driver looks up a vproc means/processor node translation in the translation table and sends the message to a designated processor node and the designated processor node routes it locally to an appropriate mailbox in the particular vproc means.

11. A computer system as recited in claim 7, further comprising a clustering of processor nodes connected to a common set of data storage devices, wherein the processor nodes are cross-connected to the common set of data storage devices.

12. A computer system as recited in claim 11, wherein each vproc means is connected to at least one virtual disk comprising one or more memory regions of the data storage devices that are grouped together, wherein the virtual disk is accessible by any processor node within a clique, thereby allowing the vproc means to be started and executed on any processor node within a clique.

13. A method of communicating between a plurality of threads executing in a parallel computer system comprising a plurality of processor nodes after a processor node failure, the method comprising the steps of:

(a) utilizing a plurality of vprocs to perform multiple threads simultaneously, wherein each vproc comprises an operating environment in one of the processor nodes for performing one of said multiple threads, each one of said multiple threads being a stream of instructions executed on behalf of a task, wherein each one of said multiple threads access data on a different data storage device simultaneously with the rest of said multiple threads;

(b) moving a vproc from a first processor node to a second processor node after the first processor node has failed;

(c) updating a vproc/processor node translation table to reflect the movement of the vproc from the first processor node to the second processor node, so that any messages are routed to the vproc in the second processor node;

(d) storing information from the vproc on a virtual disk comprised one or more memory regions of the data storage devices that are grouped together, wherein the virtual disk is accessible by any processor node within a clique;

(e) moving the vproc to a processor node in the clique; and

(f) restarting a thread performed by the moved vproc using information retrieved from the virtual disk.

14. A method as recited in claim 13, comprising the steps of:

(1) adding an available processor node to said parallel computer system; and

(2) selectively moving vprocs from other processor nodes to the available processor node until the vprocs are optimally distributed between the processor nodes.

15. A method of executing instructions in a parallel computer system comprised of a plurality of processor nodes, each of the processor nodes having a plurality of data storage devices connected thereto, and an interconnect network, coupled to the processor nodes, for transmitting messages between processor nodes, the method comprising:

(a) utilizing a plurality of vprocs to perform multiple threads simultaneously, wherein each vproc comprises an operating environment in one of the processor nodes for performing one of said multiple threads, each one of said multiple threads being a stream of instructions executed on behalf of a task, wherein each one of said multiple threads access data on a different data storage device simultaneously with the rest of said multiple threads; and

(b) assigning each of the vprocs to one of the processor nodes at system startup, further comprising creating and storing a translation table identifying locations of vprocs and processor nodes on the interconnect network at the system startup.

16. A method as recited in claim 15, further comprising the step of transmitting messages between vprocs, wherein each of the vprocs includes one or more mailboxes, such that when a message is sent to a mailbox at a particular vproc means, an interconnect network driver looks up a vproc/processor node translation in the translation table and sends the message to a designated processor node and the designated processor node routes it locally to an appropriate mailbox in the particular vproc means.

17. A method as recited in claim 15, further comprising the step of clustering processor nodes into a clique connected to a common set of data storage devices, wherein the processor nodes are cross-connected to the common set of data storage devices.

18. A method as recited in claim 17, wherein each vproc is connected to at least one virtual disk comprising one or more memory regions of the data storage devices that are grouped together, and the virtual disk can be accessed by a processor node within a clique, thereby allowing the vproc to be started and executed on any processor node within a clique.

19. A method of executing instructions in a parallel computer system comprised of a plurality of processor nodes, each of the processor nodes having a plurality of data storage devices connected thereto, and an interconnect network, coupled to the processor nodes, for transmitting messages between processor nodes, the method comprising:

(a) utilizing a plurality of vprocs to perform multiple threads simultaneously, wherein each vproc comprises an operating environment in one of the processor nodes for performing one of said multiple threads, each one of said multiple threads being a stream of instructions executed on behalf of a task, wherein each one of said multiple threads access data on a different data storage device simultaneously with the rest of said multiple threads; and

(b) assigning the vprocs to processor nodes at system startup, including the steps of creating and storing a translation table identifying locations of vprocs and processor nodes on the interconnect network at system startup, and providing updates to the translation table when movement of the vprocs between processor nodes occur during runtime.

20. A method as recited in claim 19, further comprising the step of transmitting messages between vprocs, wherein each vproc includes one or more mailboxes, such that when a message is sent to a mailbox at a particular vproc, an interconnect network driver looks up a vproc/processor node translation in the translation table and sends the message to a designated processor node and the designated processor node routes it locally to an appropriate mailbox in the particular vproc.

21. A method as recited in claim 19, further comprising the step of clustering processor nodes in a clique connected to a common set of data storage devices, wherein the processor nodes are cross-connected to the common set of data storage devices.

22. A method as recited in claim 21, wherein each vproc is connected to at least one virtual disk comprising one or more memory regions of the data storage devices that are grouped together, wherein the virtual disk is accessible by any by a processor node within a clique, thereby allowing the vproc means to be started and executed on any processor node within a clique.