Load nvidia-uvm at boot time and install nvidia-persistenced as syste…

…m service on GPU instances

When running on a GPU instance, this commit loads kernel module of Nvidia unified virtual memory by default and install Nvidia persistence daemon as a system service.
Nvidia unified virtual memory makes it easy to use memory on both CPU and GPU.
Nvidia persistence daemon keeps GPU initialized, therefore shorten application startup latency.
See reference:
Nvidia-uvm: https://developer.nvidia.com/blog/unified-memory-cuda-beginners/
Nvidia persistence daemon: https://docs.nvidia.com/deploy/driver-persistence/index.html

Signed-off-by: Hanwen <[email protected]>

CHANGELOG.md

@@ -24,6 +24,8 @@ This file is used to list changes made in each version of the AWS ParallelCluste
 - Add log rotation support for ParallelCluster managed logs.
 - Track head node memory and root volume disk utilization using the `mem_used_percent` and `disk_used_percent` metrics collected through the CloudWatch Agent.
 - Enforce the DCV Authenticator Server to use at least `TLS-1.2` protocol when creating the SSL Socket.
+- Load kernel module [nvidia-uvm](https://developer.nvidia.com/blog/unified-memory-cuda-beginners/) by default.
+- Install [Nvidia persistence daemon](https://docs.nvidia.com/deploy/driver-persistence/index.html) as a system service.
 
 **CHANGES**
 - Upgrade Slurm to version 23.02.1.

cookbooks/aws-parallelcluster-config/files/default/nvidia/nvidia.conf

@@ -0,0 +1 @@
+nvidia-uvm

cookbooks/aws-parallelcluster-config/recipes/base.rb

@@ -39,6 +39,9 @@
   action :configure
 end
 
+# Configure Nvidia driver
+include_recipe "aws-parallelcluster-config::nvidia"
+
 # EFA runtime configuration
 efa 'Configure system for EFA' do
   action :configure

cookbooks/aws-parallelcluster-config/recipes/nvidia.rb

@@ -0,0 +1,41 @@
+# frozen_string_literal: true
+
+#
+# Cookbook:: aws-parallelcluster
+# Recipe:: nvidia
+#
+# Copyright:: 2013-2021 Amazon.com, Inc. or its affiliates. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License"). You may not use this file except in compliance with the
+# License. A copy of the License is located at
+#
+# http://aws.amazon.com/apache2.0/
+#
+# or in the "LICENSE.txt" file accompanying this file. This file is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES
+# OR CONDITIONS OF ANY KIND, express or implied. See the License for the specific language governing permissions and
+# limitations under the License.
+
+if graphic_instance? && nvidia_installed?
+  # Load kernel module Nvidia-uvm
+  kernel_module 'nvidia-uvm' do
+    action :load
+  end
+  # Make sure kernel module Nvidia-uvm is loaded at instance boot time
+  cookbook_file 'nvidia.conf' do
+    source 'nvidia/nvidia.conf'
+    path '/etc/modules-load.d/nvidia.conf'
+    owner 'root'
+    group 'root'
+    mode '0644'
+  end
+  # Make sure nvidia_persistenced is installed as a system service
+  bash 'nvidia.run advanced' do
+    cwd '/usr/share/doc/NVIDIA_GLX-1.0/samples'
+    user 'root'
+    group 'root'
+    code <<-NVIDIA
+      tar -xf nvidia-persistenced-init.tar.bz2
+      ./nvidia-persistenced-init/install.sh
+    NVIDIA
+  end
+end

test/recipes/controls/aws_parallelcluster_config/nvidia_spec.rb

@@ -39,6 +39,26 @@
   end
 end
 
+control 'tag:config_nvidia_uvm_and_persistenced_on_graphic_instances' do
+  only_if do
+    !(os_properties.centos7? && os_properties.arm?) &&
+      !instance.custom_ami? && instance.graphic?
+  end
+
+  describe kernel_module('nvidia_uvm') do
+    it { should be_loaded }
+  end
+
+  describe file('/etc/modules-load.d/nvidia.conf') do
+    its('content') { should include("uvm") }
+  end
+
+  describe service('nvidia-persistenced') do
+    it { should be_enabled }
+    it { should be_running }
+  end
+end
+
 control 'tag:config_gdrcopy_disabled_on_non_graphic_instances' do
   only_if do
     !(os_properties.centos7? && os_properties.arm?) &&