Back to results
US 12,112,200 B2Granted
Pipeline parallel computing using extended memory
AssigneeInternational Business Machines Corporation
InventorsAbdullah Kayi, Tayfun Gokmen
Life of this patent
Estimating the term…
Abstract
A system comprises compute nodes distributed over a network and configured to perform a pipeline parallel process. The system also comprises an extended memory comprising a global virtual address space which is shared by the compute nodes. The extended memory is configured to enable the compute nodes to exchange data over the network when the compute nodes perform the pipeline parallel process.
Claims20
A system, comprising:
- compute nodes operatively coupled to a communications network, each compute node comprising processor circuitry and a local memory;
- and
- wherein portions of the local memory of each compute node are pooled to provide an extended memory which comprises a global virtual address space which is shared by the processor circuitry of the compute nodes;
- wherein the processor circuitry of the compute nodes communicate over the communications network to perform a pipeline parallel computation process and utilize the extended memory to exchange data over the communications network to perform the pipeline parallel computation process;
- and
- wherein in performing the pipeline parallel computation process, data generated by the processing circuitry of a first compute node performing a respective first computation is stored in the extended memory and accessed and utilized by the processing circuitry of a second compute node to perform a respective second computation.
3 more elements
A computer program product, comprising:
- one or more non-transitory computer readable storage media, and program instructions collectively stored on the one or more non-transitory computer readable storage media, the program instructions comprising:
- program instructions to configure compute nodes to perform a pipeline parallel computation process, wherein the compute nodes are distributed over a communications network, each compute node comprising a hardware accelerator device and a local memory;
- and
- program instructions to configure an extended memory comprising a global virtual address space which is shared by the compute nodes, the extended memory comprising a virtual memory pool which comprises at least a portion of each local memory of each compute node;
- wherein the extended memory is configured to enable the compute nodes to exchange data over the communications network when the compute nodes perform the pipeline parallel computation process;
- and
- wherein in performing the pipeline parallel computation process, data generated by a first compute node performing a respective first computation is stored in the extended memory and accessed and utilized by a second compute node to perform a respective second computation.
4 more elements
A method comprising:
- configuring compute nodes to perform a pipeline parallel computation process, wherein the compute nodes are distributed over a communications network, each compute node comprising a hardware accelerator device and a local memory;
- and
- configuring an extended memory which comprises a global virtual address space that is shared by the compute nodes, the extended memory comprising a virtual memory pool which comprises at least a portion of each local memory of each compute node;
- wherein the extended memory is configured to enable the compute nodes to exchange data over the communications network when the compute nodes perform the pipeline parallel computation process;
- and
- wherein in performing the pipeline parallel computation process, data generated by a first compute node performing a respective first computation is stored in the extended memory and accessed and utilized by a second compute node to perform a respective second computation.
3 more elements
Dependencies are inferred from the claim text.