SJTU-HPC基本操作

官方文档

LMOD软件管理

https://lmod.readthedocs.io/en/latest/010_user.html

交大hpc使用lmod来管理用户软件，每一次重新登陆都会重置module，需要将软件重新load。
常用指令：

ml # list these loaded modules
ml miniconda3 # load the miniconda3 module
ml -miniconda3 # unload the miniconda3 module
module spider # inspect the possible modules that can be loaded
module spider cuda # show available cuda versions

oh-my-zsh更好的shell

hpc默认已安装zsh, 通过omzsh的安装脚本进行安装，并安装插件

export CHSH=no  # 避免它自动切换默认 shell（HPC 通常不允许）
sh -c "$(curl -fsSL https://raw.githubusercontent.com/ohmyzsh/ohmyzsh/master/tools/install.sh)"
git clone https://github.com/zsh-users/zsh-syntax-highlighting.git \
  ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/zsh-syntax-highlighting
git clone https://github.com/zsh-users/zsh-autosuggestions \
  ${ZSH_CUSTOM:-~/.oh-my-zsh/custom}/plugins/zsh-autosuggestions
git clone --depth 1 https://github.com/junegunn/fzf.git ~/.fzf
~/.fzf/install
git clone https://github.com/psprint/zsh-navigation-tools ~/.zsh-navigation-tools

## vim add configuration in `.zshrc`
plugins=(
  git
  zsh-navigation-tools
  rails
  ruby
  zsh-syntax-highlighting
  zsh-autosuggestions
)
source ${(q-)ZSH_CUSTOM:-$ZSH/custom}/plugins/zsh-syntax-highlighting/zsh-syntax-highlighting.zsh

## :wq

nvim 更好的编辑器

因为在服务器上没有sudo权限所以需要手动编译

首先使用conda创建编译环境

conda create -n nvim-build cmake libtool gettext curl unzip ninja gcc gxx -y -c conda-forge

clone & build

git clone https://github.com/neovim/neovim.git
cd neovim
git checkout stable
make CMAKE_BUILD_TYPE=Release

Tricks

交互式计算资源

某些情况下需要进行并行编译，编译需要耗费大量资源，并不一定可以在登录节点使用：

在这种情况下使用提交作业的方式较为低效，可以申请交互式计算资源

srun -p cpu -n 4 --pty /bin/bash

example:

(base) chenyulin@pilogin2 ~> srun -p cpu -n 4 --pty /bin/bash
srun: job 43090290 queued and waiting for resources
srun: job 43090290 has been allocated resources
(base) chenyulin@cas242 ~> ls
condalist.txt  dinov2  log  test.slurm
(base) chenyulin@cas242 ~> cd dinov2        
(base) chenyulin@cas242 ~/dinov2> conda activate dinov2
(dinov2) chenyulin@cas242 ~/dinov2> export CC=$CONDA_PREFIX/bin/gcc                       main
export CXX=$CONDA_PREFIX/bin/g++
(dinov2) chenyulin@cas242 ~/dinov2> which g++                                             main
~/.conda/envs/dinov2/bin/g++
(dinov2) chenyulin@cas242 ~/dinov2> vim conda_cyl.yaml                                    main
(dinov2) chenyulin@cas242 ~/dinov2> conda env update -f conda_cyl.yaml                    main
Retrieving notices: ...working... done
Channels:
 - xformers
 - pytorch
 - nvidia
 - conda-forge
 - defaults
 - nvidia/label/cuda-11.7.0
Platform: linux-64
Collecting package metadata (repodata.json): done
Solving environment: done
Installing pip dependencies: |

文件传输

本地向云端传输

使用hpc studio传输
https://studio.hpc.sjtu.edu.cn/pun/sys/dashboard/files/fs//lustre/home/acct-umjbyy/chenyulin/Data

云端之间互相传输

例如拷贝conda环境

cp -r /lustre/home/acct-umjbyy/chenyulin/.conda /dssg/home/acct-umjbyy/chenyulin/