- Some notes on using our cluster:
- Please use SLURM to submit your tasks. Ask your labmates to get a copy of slurm script.
- Please sbatch your job.slurm through master node, but do not run jobs directly on master node.
- If you need to load ParaView, do not load gcc module. ParaFoam is invalid, enter paraview directly.
- If the job ends abnormally or fails to start, login compute nodes (10.11.13.107~110) to check whether there are zombie processes (Enter top to see the ongoing command). Use pkill -u user_name -f command_name to cancel.
- In OpenFOAM cases, 5,000~10,000 grids per core are preferred. If the computational domain is decomposed into too many processors with less than 1,000 grids per core, the improvement of efficiency is seldom and even worse.
- The computational speed decreases slightly in mixing node using. For example, if you need to run a 128-core task on node01, while another 60-core task is running also on node01, the task speed would decrease by 10%~20% compared with exclusive use. So please make full use of each node!
- Cross-node would not cause extra deceleration in our server. If you need to run a long task with 2 or 3 nodes allocated, please let the advisor know.
- Login to JI server
- Download VPN app from https://202.121.180.2:4430/. In the VPN, the server address fill in: https://202.121.180.2:4430/
- Fill in user name and log in VPN
- vpn连接成功后,请ssh 访问cluster 地址10.11.11.2,用户名dezhi.zhou
- Login to JIM server
username password Ip address for the server dezhi_zhou Server 1: 10.11.13.37
Server 2: 10.11.13.69
Server 3: 10.11.19.27
root jim 请注意
- dezhi_zhou已经加入sudo组
- 已禁止root用户远程ssh密码登陆,请勿更改策略。
- 所有SSD空间已经分配到根目录,所有HDD空间已经分配到data目录。
- 服务器目前配置为2SSD+3HDD, 2SSD 已做raid 0 , 3HDD 已做raid 5。 Raid 0 相比单SSD性能有较大提升,但是不提供数据冗余,请做好日常的数据备份。
- 后台运行程序 nohup ./a.out
- 查看job:jobs
- access remote repository through SSH
- Generate SSH key by “ssh-keygen -t rsa -C “your email””
- I will produce a .pub file. You should copy the content in this file to the github website (find your account – setting – SSH key).
- eval `ssh-agent -s`
- Then “ssh-add fileName”, where the file is the one you produced in the previous step
- Then test with “ssh -T git@github.com”
- “git remote rm origin”
- “git remote add origin link”, where you can copy the link in the repisitory on the github website (the SSH link).
- Notes on install OF in Siyuan HPC
-
Source this before compling or running OF on hpc:”’#module load python #may need when compling, no need in runningmodule load intel-oneapi-mpi/2021.6.0 #better specify a fixed versionmodule load intel-oneapi-compilers/2022.2.1export CPATH=$CPATH:/dssg/home/acct-umjzdz/umjzdz-zzw/.conda/envs/cantera_OF/include #may need when compilingsource $HOME/OpenFOAM/OpenFOAM-6/etc/bashrc”’
-
Change the etc/bashrc configuration to ‘ICC’ as compiler and ‘INTELMPI’ as mpi,in etc/bashrc, add export MPI_ROOT=$I_MPI_ROOT
-
In etc/config.sh/compiler, add”’Icc)export WM_CC=’icc’export WM_CXX=’icpc’;;”’after the setting of Clang
- If show UCX error when using mutiple nodes, please add this to your slurm script: export UCX_NET_DEVICES=mlx5_0:1
-