Ubuntu安装PBG并提供GPU支持

Ubuntu安装PBG

硬件/软件配置

• ubuntu 18.04
• nvidia-smi 440.100

使用docker进行安装

可根据需求以root用户权限或以非root用户权限安装Pytorch和PBG。

• 配置pip源

  cd ~
  mkdir .pip
  cd .pip
  touch pip.conf
  echo "[global]" >> ~/.pip/pip.conf
  echo "index-url = https://pypi.tuna.tsinghua.edu.cn/simple" >> ~/.pip/pip.conf
  echo "[install]" >> ~/.pip/pip.conf
  echo "trusted-host = https://pypi.tuna.tsinghua.edu.cn" >> ~/.pip/pip.conf
  cd /home/MyWorkspace
  

• 根据驱动满足的最大cuda版本来查找合适版本的pytorch，不需要单独安装cuda
  • Previous PyTorch Versions
  • 使用pip安装pytorch
• 下载PBG源码进行编译安装，不需要单独安装cuda

  wget https://github.com/facebookresearch/PyTorch-BigGraph/archive/refs/heads/main.zip
  unzip main.zip
  rm main.zip
  cd PyTorch-BigGraph-main
  PBG_INSTALL_CPP=1 pip install .
  

  若不使用GPU，则可以使用pip install .进行安装
  根据需要，使用PBG提供的案例mv torchbiggraph/examples ../pbg-examples和删除安装包cd .. && rm -rf PyTorch-BigGraph-main。
  PBG的python库和可执行命令分别被安装在**/site-packages和**/bin中。

使用Docker进行部署PBG

• 打包镜像

  sudo docker commit -m "pip install pbg with cuda:10.2 and pytorch:1.12.0" container_id pbg_single:v1.0
  

• 编写docker-compose.yml以部署单个容器

  version: '3'
  services:
  	dgl_cluster:
  		image: pbg_single:v1.0
          user: "pbg"
          working_dir: /home/pbg/MyWorkspace
          volumes:
          	- ./pbg:/home/pbg/MyWorkspace
          shm_size: '32g'
          storage_opt:
          	size: '256g'
          deploy:
          	resources:
          	    limits:
            			cpus: '8.0'
            			memory: 16G
          		reservations:
          			devices:
          				- driver: nvidia
          				  count: "all"
          				  capabilities: [gpu]
          stdin_open: true
          tty: true
          command: /bin/bash -c "/bin/bash /home/pbg/.ssh/sshd.sh && /bin/bash"
  

• 使用docker-compose部署容器

  sudo docker-compose up -d
  

使用Docker部署分布式PBG

使用Docker部署DistDGL

• 安装sshd并配置ssh（在安装PBG之前或之后均可）
• 打包镜像

  sudo docker commit -m "pip install pbg with cuda:10.2 and pytorch:1.12.0 and sshd" container_id pbg_single:v1.0
  

• 部署单个容器

  version: '3.5'
  services:
  	pbg_cluster:
  		image: pbg_single:v1.0
          user: "pbg"
          working_dir: /home/pbg/MyWorkspace
          volumes:
          	- ./pbg:/home/pbg/MyWorkspace
          shm_size: '32g'
          storage_opt:
          	size: '256g'
          deploy:
          	resources:
          	    limits:
            			cpus: '8.0'
            			memory: 16G
          		reservations:
          			devices:
          				- driver: nvidia
          				  count: "all"
          				  capabilities: [gpu]
          stdin_open: true
          tty: true
          command: /bin/bash -c "/bin/bash /home/pbg/.ssh/sshd.sh && /bin/bash"
  

• 部署多个镜像

  sudo docker-compose up --scale pbg_cluster=2 -d