Resilient message passing in applications executing separate threads in a plurality of virtual compute nodes

• US 9,384,052 B2
• Filed: 12/20/2010
• Issued: 07/05/2016
• Est. Priority Date: 12/20/2010
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Dalal, Danny, Mura, Marco
• Primary Examiner(s): Wai, Eric C
• Application Number:
  US12/972,545
• Publication Number:
  US 20120159505A1
• Time in Patent Office: 2,024 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 2209/547 Messaging middleware
• CPC Class Codes: G06F 9/5027 the resource being a machine, e.g. CPUs, Servers, Terminals

Abstract

A message passing system may execute a parallel application on multiple compute nodes. Each compute node may perform a single workload on at least two physical computing resources. Messages may be passed from one compute node to another, and each physical computing resource assigned to a compute node may receive and process the messages. In some embodiments, the compute nodes may be virtualized so that a message passing system may only detect a single compute node and not the multiple underlying physical computing resources.

Claims

1. A system comprising:

a plurality of virtual compute nodes, each of the virtual compute nodes comprises at least two physical computing resources, each of the physical computing resources perform identical computing tasks for a virtual compute node;

an application manager that is configured to receive an application to execute, said application having a plurality of execution threads, each of said execution threads being operable on a different virtual computing node in the plurality of virtual compute nodes, said application further configured to use a message passing system to pass messages from one execution thread to another; and

an application execution environment that exposes said virtual compute nodes to the application, said application execution environment configured to;

receive, from the application manager;

a first execution thread of the parallel processing application; and

an identity of a first virtual compute node in the plurality of virtual compute nodes, the first virtual compute node comprises the at least two physical computing resources that are to execute the first execution thread;

receive a first message from a second virtual compute node executing a second execution thread of the application, the first message being addressed to the first virtual compute node;

provide the first message to the at least two physical computing resources of the first virtual compute node, each of the at least two physical computing resources executing the first execution thread for the application;

receive a second message from a first physical computing resource in the at least two physical computing resources, the second message based upon the first message, the second message having an address for a third virtual compute node; and

transfer the second message to the third virtual compute node.

2. The system of claim 1, said application execution environment further configured to:

detect that a first physical computing resource of the first virtual compute node is unavailable; and

remove said first physical computing resource from the first virtual compute node such that the first physical computing resource is no longer capable of receiving messages directed to the first virtual compute node.

3. The system of claim 1, said first virtual compute node having a computing resource manager that is configured to:

receive the first message for said first virtual compute node and pass said first message to the first application thread executing on said at least two physical computing resources.

4. The system of claim 3, said computing resource manager further configured to:

receive a third message from a second of said at least two physical computing resources and fail to transmit said third message, said third message having a work payload that is identical to a work payload of said second message.

5. The system of claim 4, further comprising at least one non-virtualized compute node.

10. The system of claim 1, the physical computing resources being respective server computing devices.

6. The system of claim,

said application manager further configured to:

determine a number of virtual compute nodes to be assigned to said application;

identify physical computing resources to execute an execution thread and configure a computing resource manager for said physical computing resources; and

cause said execution threads to be started on each of said physical computing resources.

7. The system of claim 6, said computing resource manager being executed on each of said physical computing resources.

8. The system of claim 6, said computing resource manager being executed on a first physical computing resource and not on one of said at least two physical computing resources of the first virtual compute node.

9. The system of claim 8, said computing resource manager being executed on a physical computing resource that is not one of said at least two physical computing resources.

11. A method for operating a first virtual compute node in a message passing environment for executing a parallel processing application, said method comprising:

receiving, from an application manager;

a first execution thread of the parallel processing application; and

an identity of at least two physical computing resources to execute the first execution thread, the first virtual compute node comprises the at least two physical computing resources;

receiving a first message from a second virtual compute node executing a second execution thread of the parallel processing application, the first message being addressed to said first virtual compute node;

providing the first message to the at least two physical computing resources of the first virtual compute node, each of said at least two physical computing resources executing the first execution thread for said parallel processing application;

receiving a second message from a first physical computing resource in the at least two physical computing resources, the second message based upon the first message, said second message having an address for a third virtual compute node; and

transferring said second message to said third virtual compute node.

12. The method of claim 11 further comprising:

receiving a third message from a second physical computing resource in the at least two physical computing resources, said third message having an address for said third virtual compute node;

determining that said third message and said second message are identical; and

failing to transmit said third message based upon the third message and the second message being identical.

13. The method of claim 11 further comprising:

receiving a third message from a second physical computing resource in the at least two physical computing resources, said third message having an address for said third compute node;

determining that said third message and said second message are different;

identifying that one of the first physical computing resource or the second computing resource has failed based upon said third message and said second message being different; and

removing the one of the first physical computing resource or the second computing resource that has failed from the first virtual compute node.

14. The method of claim 13, the first physical computing resource being a first server computing device and the second physical computing resource being a second server computing device.

15. A system comprising:

multiple physical computing resources;

an application manager that is configured to receive an application to execute, said application having a plurality of workflow portions, each of said workflow portions being executable on a different virtual compute node amongst a first set of virtual compute nodes, said application further configured to use a message passing system to pass messages from one workflow portion to another;

an application execution environment that is configured to;

expose the first set of virtual compute nodes to the application, each of said virtual compute nodes executing a workflow portion of said application;

transmit communications from a first virtual compute node in the first set of virtual compute nodes to a second virtual compute node in the first set of virtual compute nodes;

for each of said virtual compute nodes;

provide at least two physical computing resources, each of said physical computing resources executing an identical workflow portion; and

forward communications received by a virtual compute node to each of said physical computing resources associated with said virtual compute node.

16. The system of claim 15, one of said physical computing resources being designated a primary physical computing resource and a second of said physical computing resources being designated a duplicate physical computing resource.

17. The system of claim 15, said application execution environment further configured to:

detect that a third physical computing resource has become unavailable; and

remove said third physical computing resource from receiving communications.

18. The system of claim 15, said application execution environment further configured to:

query each of said physical computing resources to determine whether or not each physical computing resource received a message, and

send said message when said message had not been received.

19. The system of claim 18, said application execution environment further configured to:

detect a failed communication by;

setting a timer at a first communication attempt for a first message; and

detecting expiration of said timer prior to detecting a successful receipt of said first message.

20. The system of claim 15, the multiple physical computing resources being servers in a data center.