Message passing among interdependent parallel processes using a shared memory
View Patent ↗Message passing is provided among a plurality of interdependent parallel processes using a shared memory. Inter-process communication among a plurality of interdependent processes executing on a plurality of compute nodes is performed by obtaining a message from a first process for a second process; and storing the message in a memory location of a Peripheral Component Interconnect Express (PCIE)-linked storage device, wherein the second process reads the memory location to obtain the message. The message is optionally persistently stored in the PCIE-linked storage device for an asynchronous checkpoint until the message is no longer required for an asynchronous restart.
1. A method for inter-process communication among a plurality of interdependent processes executing on a plurality of compute nodes, comprising:
obtaining a message by means of a Peripheral Component Interconnect Express (PCIE) network interconnecting a plurality of compute nodes, wherein said message is from a first process executing on a first compute node of the plurality of compute nodes for a second process executing on a distinct second compute node of the plurality of compute nodes; and
storing said message in a predefined persistent memory location of a PCIE-linked persistent storage device that is distinct from said first compute node and said second compute node, wherein said second process reads said predefined persistent memory location using a PCIE communication to obtain said message from said predefined persistent memory location.
2. The method of claim 1 , wherein said first process employs a Message Passing Interface (MPI)-Send function of an MPI application programming interface (API) to send said message.
3. The method of claim 2 , wherein said second process employs a MPI-Receive function of said MPI API to read said message.
4. The method of claim 1 , wherein said first process and said second process employ a PCIE bus on a respective compute node where said first process and said second process execute.
5. The method of claim 1 , wherein said PCIE-linked storage device employs a PCIE bus on said PCIE-linked storage device.
6. The method of claim 1 , wherein said predefined persistent memory location comprises a unique value for said first process and said second process.
7. The method of claim 1 , wherein said predefined persistent memory location is locked until said second process reads said predefined persistent memory location to obtain said message.
8. The method of claim 1 , wherein said predefined persistent memory location stores an offset to a second memory location where at least a portion of said message is stored.
9. The method of claim 1 , wherein said message is persistently stored in said PCIE-linked persistent storage device for an asynchronous checkpoint.
10. The method of claim 9 , wherein said message is persistently stored in said PCIE-linked persistent storage device until said message is no longer required for an asynchronous restart.
11. The method of claim 1 , wherein said first and second processes comprise portions of a High Performance Computing (HPC) workload in an enterprise environment.
12. A non-transitory machine-readable recordable storage medium, wherein one or more software programs when executed by one or more processing devices implement the following steps:
obtaining a message by means of a Peripheral Component Interconnect Express (PCIE) network interconnecting a plurality of compute nodes, wherein said message is from a first process executing on a first compute node of the plurality of compute nodes for a second process executing on a distinct second compute node of the plurality of compute nodes; and
storing said message in a predefined persistent memory location of a PCIE-linked persistent storage device that is distinct from said first compute node and said second compute node, wherein said second process reads said predefined persistent memory location using a PCIE communication to obtain said message from said predefined persistent memory location.
13. A system for inter-process communication among a plurality of interdependent processes executing on a plurality of compute nodes, comprising:
a memory; and
at least one processing device configured to:
obtain a message by means of a Peripheral Component Interconnect Express (PCIE) network interconnecting a plurality of compute nodes, wherein said message is from a first process executing on a first compute node of the plurality of compute nodes for a second process executing on a distinct second compute node of the plurality of compute nodes; and
store said message in a predefined persistent memory location of a PCIE-linked persistent storage device that is distinct from said first compute node and said second compute node, wherein said second process reads said predefined persistent memory location using a PCIE communication to obtain said message from said predefined persistent memory location.
14. The system of claim 13 , wherein said first process employs a Message Passing Interface (MPI)-Send function of an MPI application programming interface (API) to send said message and wherein said second process employs a MPI-Receive function of said MPI API to read said message.
15. The system of claim 13 , wherein said first process and said second process employ a PCIE bus on a respective compute node where said first process and said second process execute.
16. The system of claim 13 , wherein said PCIE-linked storage device employs a PCIE bus on said PCIE-linked storage device.
17. The system of claim 13 , wherein said predefined persistent memory location comprises a unique value for said first process and said second process.
18. The system of claim 13 , wherein said predefined persistent memory location stores an offset to a second memory location where at least a portion of said message is stored.
19. The system of claim 13 , wherein said message is persistently stored in said PCIE-linked persistent storage device for an asynchronous checkpoint.
20. The system of claim 13 , wherein said message is persistently stored in said PCIE-linked persistent storage device until said message is no longer required for an asynchronous restart.