Find the Enable FPGA Usage property and select the
NodeManagers that can provide FPGA devices to YARN applications that request
them.
Use the Path to FPGA (aocl) tool property to specify the
full local path to the Intel aocl tool installed on the applicable nodes.
Use the Allowed FPGA devices to specify the FPGA devices
which can be managed by YARN NodeManager.
Valid values are the following:
Auto: Default value. All available FPGA devices are
allowed.
Comma separated list: List the minor number of the allowed FPGAs.
For example: 0,1
Use the List of available FPGA devices property to
manually specify the available FPGA devices.
Provide the value in the following format:
deviceA/major_number:minor_number,deviceB/major_number:minor_number
For example: acl0/238:0,acl1/238:1
The acl numbers are displayed using the aocl diagnose
command.
The major and minor device numbers can usually be determined by listing the
device files under /dev. Every card has a corresponding
device file. The ls -l command shows both numbers.
Use the FPGA device major number property to provide the
major device number of the FPGA card.
This property is used in the container-executor.cfg
file.
Use the FPGA initializer script property to provide the
path to the shell script that sets up the environment for the FPGA device.
The initializer script sets up various environment variables, and sources
other scripts allowing the aocl tool to run properly. The
contents of this script depends heavily on the installation. For example, in
case of Intel Processing Accelerator Card (PAC), it is necessary to source
init_env.sh and init_opencl.sh.
The following environment variables are possibly required to be exported:
AOCL_BOARD_PACKAGE_ROOT
CL_CONTEXT_COMPILER_MODE_INTELFPGA
Please consult the vendor’s documentation for further details.
Click Save Changes.
Restart the affected NodeManagers.
FPGA support is enabled and configured.Request FPGA resource by specifying yarn.io/fpga as a
resource.