I've been using torque as a job scheduler for numerical workstations for many years, but now that I'm charged, I've introduced the free software Slurm as an alternative.
For the introduction of Slurm to Ubuntu 16.04 LTS, the following article https://qiita.com/JeJeNeNo/items/c545e72373d4d1deb36a It is carefully summarized in, and installation on 18.04 LTS can be done almost like this. For the detailed procedure, see the above article, but I will summarize some changes. If you want to know from 1, I recommend you to read the above article first.
The procedure up to installation is the same as the above article, so it will be omitted. Before installing Slurm, it was necessary to install the software MUNGE required to start Slurm, and it seems that 16.04 LTS was difficult. Fortunately, 18.04LTS can be installed with apt install without any problems.
sudo apt install munge
It seems that Slurm can also be installed with apt install, but considering the later settings, I compiled and installed the latest version from the source code as in the above article. Since I did not set the prefix, it will be installed in "/ usr / local" according to the default setting.
./configure
make
sudo make install
This is almost the same as the procedure in the above article, but some changes and additions. In particular,
After that, follow the above article to open the port, set it to systemd, and start it.
It is organized on the following site. https://www.j-focus.jp/user_guide/ug0004000000/ What you should remember for the time being -Job execution
sbatch **.sh
-Check job status
squece
・ Job forced termination
scancel ***
So, how to write a batch file worked if I modified the sample below for myself. https://www.j-focus.jp/user_guide/ug0004040000/#ug0004040000
Installation work and usage are easier than torque, and it's free to complain, but the manual on the official website was hard to read and closed.
Recommended Posts