English

Índice

1. Log in to the cluster
   1. Login nodes (logins)
   2. Password change
   3. Access from/to the outside
2. Directories and file systems
   1. Basic directories under FEFS
   2. Basic directories under GPFS
3. Running jobs
   1. Submit to queues
   2. Interactive jobs

Log in to the cluster ↩

IMPORTANT

The accounts are for personal and non-transferable usage. If the project requires someone else's access to the machine or any increase in the assigned resources, the project manager will be in charge to make this type of request.

Login nodes (logins) ↩

Cluster	Login nodes
MareNostrum 4 (GP)	mn1.bsc.es mn2.bsc.es mn3.bsc.es
CTE-ARM	armlogin1.bsc.es armlogin2.bsc.es

All connections must be done through SSH (Secure SHell), for example:

 mylaptop$> ssh {username}@armlogin1.bsc.es
 mylaptop$> ssh {username}@armlogin2.bsc.es

Password change ↩

For security reasons, you must change the first password.

To change your password, you have to login at Storage (Data Transfer machine):

mylaptop$> ssh {username}@dt01.bsc.es

using the same username and password than in the cluster. Then, you have to use the 'passwd' command.

The new password will become effective after about 10 minutes.

Access from/to the outside ↩

The login nodes are the only nodes accessible from the outside, but no connections are allowed from the cluster to the outside world for security reasons.

All file transfers from/to the outside must be executed from your local machine and not within the cluster:

Example to copy files or directories from MN4 to an external machine:

 mylaptop$> scp -r {username}@dt01.bsc.es:"MN4_SOURCE_dir" "mylaptop_DEST_dir"

Example to copy files or directories from an external machine to MN4:

 mylaptop$> scp -r "mylaptop_SOURCE_dir" {username}@dt01.bsc.es:"MN4_DEST_dir"

Directories and file systems ↩

There are different partitions of disk space. Each area may have specific size limits and usage policies.

Basic directories under FEFS ↩

FEFS (Fujitsu Exabyte File System) is a scalable cluster file system based on Lustre with high reliability.

The FEFS directories/file systems available and accessible from all nodes (logins and compute) are:

• /apps (softlink to /fefs/apps): over this filesystem reside applications and libraries already installed on the machine for everyday use. Users cannot write to it.

• /scratch (softlink to /fefs/scratch): it's the only filesystem intended for executions.

Basic directories under GPFS ↩

GPFS (General Parallel File System) is a distributed networked filesystem than can be accessed from all the nodes and data transfer machine (dt01.bsc.es).

The available GPFS directories and file systems are:

• /home (softlink to /gpfs/home): after login, it is the default work area where users can save source codes, scripts, and other personal data. The space quota is individual (with a relatively lessened capacity). Not recommended for run jobs; please run your jobs from /scratch instead.

• /gpfs/projects: it's intended for data sharing between users of the same group or project. All members of the group share the space quota. This filesystem is not mounted on computing nodes; you have to transfer you data and launch your jobs from /scratch.

Running jobs ↩

Submit to queues ↩

Jobs submission to the queue system have to be done through the PJM directives, for example:

To submit a job:

 $> pjsub {job_script} # similar to 'sbatch' in SLURM

To show all the submitted jobs:

 $> pjstat  # similar to 'squeue' in SLURM

To cancel a job:

 $> pjdel {job_id}  # similar to 'scancel' in SLURM

Interactive jobs ↩

Interactive sessions

Allocation of an interactive session has to be done through PJM, for example:

To request an interactive session on a compute node:

 $> pjsub --interact -L node=1

To request an interactive session on a node, specifying the 'rscgrp' (similar to 'qos' in SLURM) and duration:

$> pjsub --interact -L node=1 -L rscgrp=large -L elapse=2:00:00

Also:

$> pjsub --interact -L node=1,rscgrp=large,elapse=2:00:00