Configuring compute nodes in a parallel computer using remote direct memory access (‘RDMA’)

Patent metadata

Field	Value
Publication number	US-10474625-B2
Application number	US-201213351419-A
Country	US
Kind code	B2
Filing date	Jan 17, 2012
Priority date	Jan 17, 2012
Publication date	Nov 12, 2019
Grant date	Nov 12, 2019

Abstract

Configuring compute nodes in a parallel computer using remote direct memory access (‘RDMA’), the parallel computer comprising a plurality of compute nodes coupled for data communications via one or more data communications networks, including: initiating, by a source compute node of the parallel computer, an RDMA broadcast operation to broadcast binary configuration information to one or more target compute nodes in the parallel computer; preparing, by each target compute node, the target compute node for receipt of the binary configuration information from the source compute node; transmitting, by each target compute node, a ready message to the target compute node, the ready message indicating that the target compute node is ready to receive the binary configuration information from the source compute node; and performing, by the source compute node, an RDMA broadcast operation to write the binary configuration information into memory of each target compute node.

First claim

What is claimed is:

1. An apparatus for configuring compute nodes in a parallel computer using remote direct memory access (‘RDMA’), the parallel computer comprising an operational group comprising a plurality of compute nodes coupled for data communications via one or more data communications networks, the apparatus comprising a computer processor, a computer memory operatively coupled to the computer processor, the computer memory having disposed within it computer program instructions that, when executed by the computer processor, cause the apparatus to carry out the steps of:
initiating, by a source compute node of the operational group, an RDMA broadcast operation to broadcast binary configuration information to at least two target compute nodes of the operational group, including sending the RDMA broadcast operation message to each of the at least two target compute nodes, wherein the RDMA broadcast operation message comprises setup parameters including a size of data to be transferred during the RDMA broadcast operation, wherein the operational group comprises a Message Passing Interface (‘MPT) communicator;
preparing, by each of the at least two target computes nodes, the target compute node for receipt of the binary configuration information from the source compute node by:
changing a data communications state of the target compute node, wherein the data communication state specifies a type of data communications mode, to indicate that the target compute node is ready to receive a broadcast of data, such that the target compute node is ready to receive binary configuration information that is being broadcasted to the target compute node via an RDMA broadcast operation;
allocating memory to store the binary configuration information; and
setting a value of an expected data transfer size counter to the size of data to be transferred during the RDMA broadcast operation;

executing, by the at least two target compute nodes, a collective reduce operation, wherein the collective reduce operation sends, to the source compute node, a collective packet comprising an indication that the at least two target compute nodes are ready to receive the binary configuration information from the source compute node including an address in memory of the at least two target compute nodes to which the binary configuration information is to be written, wherein the collective reduce operation is executed in parallel by the compute nodes in the operational group;
performing, by the source compute node, the RDMA broadcast operation to write the binary configuration information into the allocated memory of each target compute node using a collective broadcast operation, including decrementing the value of the expected data transfer size counter by an amount equal to a size of the binary configuration information that was received, wherein the collective broadcast operation is executed in parallel by the compute nodes in the operational group; and
transmitting, by each of the at least two target compute nodes, a data received message to the source compute node, including transmitting a collective packet as part of a collective operation.

2. The apparatus of claim 1 wherein executing the collective reduce operation further comprises transmitting, by each node in the operational group, a collective reduce packet as part of the collective reduce operation.

3. The apparatus of claim 1 wherein transmitting, by each of the at least two compute nodes, the data received message to the source compute node comprises:
for each target compute node upon receipt of binary configuration information:
determining whether the value of the expected data transfer size counter is zero; and
responsive to determining that the value of the expected data transfer size counter is zero, transmitting the data received message to the source compute node.

4. The apparatus of claim 1 wherein the binary configuration data is a binary executable file.

5. The apparatus of claim 1 wherein MPI comprises a module of computer program instructions in the form of a parallel communications library that, when executed, is configured to support data communications amongst a plurality of compute nodes in a parallel computer.

6. A computer program product for configuring compute nodes in a parallel computer using remote direct memory access (‘RDMA’), the parallel computer comprising an operational group comprising a plurality of compute nodes coupled for data communications via one or more data communications networks, the computer program product disposed upon a computer readable medium that is not a signal, the computer program product comprising computer program instructions that, when executed, cause a computer to carry out the steps of:
initiating, by a source compute node of the operational group, an RDMA broadcast operation to broadcast binary configuration information to at least two target compute nodes of the operational group, including sending the RDMA broadcast operation message to each of the at least two target compute nodes, wherein the RDMA broadcast operation message comprises setup parameters including a size of data to be transferred during the RDMA broadcast operation, wherein the operational group comprises a Message Passing Interface (‘MPI’) communicator;
preparing, by each of the at least two target computes nodes, the target compute node for receipt of the binary configuration information from the source compute node by:
changing a data communications state of the target compute node, wherein the data communication state specifies a type of data communications mode, to indicate that the target compute node is ready to receive a broadcast of data, such that the target compute node is ready to receive binary configuration information that is being broadcasted to the target compute node via an RDMA broadcast operation;
allocating memory to store the binary configuration information; and
setting a value of an expected data transfer size counter to the size of data to be transferred during the RDMA broadcast operation;

executing, by the at least two target compute nodes, a collective reduce operation, wherein the collective reduce operation sends, to the source compute node, a collective packet comprising an indication that the at least two target compute nodes are ready to receive the binary configuration information from the source compute node including an address in memory of the at least two target compute nodes to which the binary configuration information is to be written, wherein the collective reduce operation is executed in parallel by the compute nodes in the operational group;
performing, by the source compute node, the RDMA broadcast operation to write the binary configuration information into the allocated memory of each target compute node using a collective broadcast operation, including decrementing the value of the expected data transfer size counter by an amount equal to a size of the binary configuration information that was received, wherein the collective broadcast operation is executed in parallel by the compute nodes in the operational group; and
transmitting, by each of the at least two target compute nodes, a data received message to the source compute node, including transmitting a collective packet as part of a collective operation.

7. The computer program product of claim 6 wherein executing the collective reduce operation further comprises transmitting, by each node in the operational group, a collective reduce packet as part of the collective reduce operation.

8. The computer program product of claim 6 wherein transmitting, by each of the at least two target compute nodes, the data received message to the source compute node comprises:
for each target compu

Assignees

• Aho Michael E
• Attinella John E
• Gooding Thomas M
• Mundy Michael B
• IBM

Inventors

• Aho Michael E
• Attinella John E
• Gooding Thomas M
• Mundy Michael B

Classifications

• G06F9/441

  Multiboot arrangements, i.e. selecting an operating system to be loaded · CPC title

• G06F8/63 Primary

  Image based installation; Cloning; Build to order · CPC title

• G06F9/4401

  Bootstrapping (security arrangements therefor G06F21/57) · CPC title

• G06F15/177 Primary

  Initialisation or configuration control {(processor initialisation G06F9/4405)} · CPC title