MPCDF Tutorial

Part 1 (overview)

Aim: support MPG scientist in data analysis, development etc.

MPCDF operate:

• several mid-range compute clusters,
• long-term storage tape archives,...
• remote-visualizaton infrastrcure (hardware, software)
• see webpage  (docs.mpcdf.mpg.de)

Systems:

• HPC systems (Raven, Cobra, Draco), 5 year renewal cycle
• Manage Linux clusters owned by various institutes
• RVS (remote visualization):  rvs.mpcdf.mpg.de,  purely browser based, OpenGL rendering
• Jupyter notebook service
• HPC performance monitoring service (hpc-reports.mpcdf.mpg.de)
• Data service: https://gitlab.mpcdf.mpg.de (also wikis, issues, boards) Sign-in: selfservice.mpcdf.mpg.de ; allows gues accounts
• owncloud: datashare.mpcdf.mpg.de  (sign-up: selfservice.mpcdf.mpg.de, allows guest accounts)
• Acount at mpcdf: www.mpcdf.mpg.de (For users - new users) Needs approval by director or representative

Accessing the clusters:

• Login through gateway machine  (gatezero.mpcdf.mpg.de)
• Uses 2-factor authentication; needs OTP token
• Machines to connect: raven, cobra, draco e.g. ssh raven.mpcdf.mpg.de

Software on the cluster:

• wide range of compilers & libraries
• pre-built applications
• activate through environement modules
• containers: "last resort" solution if software cannot be compiled
• Charliecloud & Singularity are available

Filesystems:

• /u   home, code, config
• /ptmp High performance FS (50-100GB/s)
• cross-mounts between the clusters (/raven /cobra /draco)

Resource manager / scheduler:

• SLURM