ARCHER2:Connection and basic commands

This page is to briefly summarize the connection and usage of the UK national supercomputing service, ARCHER2.

N.B. MAKE SURE YOU HAVE BEEN AUTHORIZED TO DO SO. Please note this page is not an official guidance. The author disclaims all responsibility for any trouble induced by the improper use of the information provided. For further details and technical supports, please visit user documentations of ARCHER2 and EPCC SAFE.

Set up the connection

An SSH key pair is needed every time the user logs in. Use the command below to generate a key pair:

$ ssh-keygen -t rsa -C your@email.com

Then follow the instructions to generate the key pair. After generating that key pair, log into your SAFE account and add the generated key pair, saved in ~/.ssh/id_rsa.pub by default, to the corresponding Login account. After the system approves the request, connect to ARCHER2 via:

$ ssh login_account@login.archer2.ac.uk

Enter the passwords for ARCHER2 account and for SSH public key in sequence. For first-time users, the default password is available in SAFE user page: In menu ‘Login accounts’, select the corresponding account, and click ‘View Login Account Password’ button on near the bottom of the screen. The default password is required to be substituted by a user-defined one during the initial login, and such change will NOT appear in the page ‘View Login Account Password’.

For the detailed, step-by-step guide, please refer: https://docs.archer2.ac.uk/user-guide/connecting/#ssh-key-pairs

General usages

The job scheduler

The Slurm job scheduler is used in ARCHER2. The job submission file should be named as jobname.slurm. Here lists some commonly used commands:

• sbatch jobname.slurm Submit the job ‘jobname’
• squeue -u $USER List the information of jobs submitted (job ID, name, status, node number)
• sinfo Check the availability of resources
• scancel jobid Kill the job with the ID number ‘jobid’.

Quality of service (QoS)

Various QoS are provided to meet different computation needs. Here lists the capacities of commonly used QoS. For full details, please find https://docs.archer2.ac.uk/user-guide/scheduler/#quality-of-service-qos.

QoS Name	CPU per node	Max nodes per job	Max walltime
Standard	128	1024	24:00

Note:

1. Slurm schedules the submitted jobs in a walltime-increasing fashion, which means jobs with larger walltime will be left in the queue for longer, so if the condition permits, increasing the number of nodes to reduce the time consumption is preferred.

CRYSTAL on ARCHER2

A good example for the job submission file is provided in this webpage. It is pasted here for reference, with the modification suggested in the same page.

A job submission script developed by myself and now used accross the group is open for download here.

Note:

1. To launch different calculations, substitute the MPPcrystal keyword in the script below with corresponding keywords:
   • MPPcrystal crystal (.d12) calculation in the massive parallel edition
   • Pcrystal crystal (.d12) calculation in a simple parallel edition
   • Pproperties parallel properties (.d3) calculation
2. Should be careful with the proper number of cores when MPPcrystal is activated. It follows the equation below:

\[n_{procs} \geq n_{tasks} = n_{r} + n_{c} \times WEIGHT\]

$n_{r}$ and $n_{c}$ are the number of k points with real or complex Fock matrices. $WEIGHT$ is the overloading of the diagonalization of the complex matrices. Its typical value is 1.5 ~ 2.5. (?)

#!/bin/bash
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=128
#SBATCH --cpus-per-task=1
#SBATCH --time=0:20:00

# Replace [budget code] below with your full project code
#SBATCH --account=[budget code]
#SBATCH --partition=standard
#SBATCH --qos=standard
#SBATCH --export=none

module load epcc-job-env
module load other-software
module load crystal

# Change this to the name of your input file
cp tio2.d12 INPUT

export FI_MR_CACHE_MAX_COUNT=0 

srun --hint=nomultithread --distribution=block:block MPPcrystal