allegro Migration

In may 2015, the upgraded allegro cluster is set online. The list of new features includes

• OS Upgrade → Debian Jessie
• new scheduler ( slurm).
• new storage technology for scratch space ( ceph).
• 3x scratch space (20T → 60T)
• more nodes
• new CPU models
• up to 512G RAM per node

The migration has happened in three phases: adoption, migration and shut down

Moving from TORQUE to slurm

You will find an overview of common commands here

There are also compatibility scripts installed so you can run your Torque-jobs directly with the new Slurm-Scheduler.

Phase I: adoption

started: 2015-04-29

closed: 2015-05-06

• torque: The old allegro cluster remains unchanged. Jobs are scheduled and processed.
• login: The new allegro login node ist opened for login.
• slurm: Users assure job requisites and start jobs on the new cluster.
• data: The file content in /home and /data/scratch is transferred to the new scratch file system allegrotmp:/data/scratch/OLD

Please note: The synchronization is repeated on a daily basis. Changes in /data/scratch/OLD might get overwritten by subsequent syncs.

Phase II: migration

started: 2015-05-06

closed: 2015-06-03

• login: The new allegro login node ist set default (i.e. allegro → allegroold, allegrotmp → allegro).
• torque: Scheduling jobs on the old allegro cluster is disabled. Scheduled jobs continue to be processed. Login remains possible via allegroold)
• nodes: Idle nodes will be moved to the new cluster
• data: The file content in allegroold:/home and allegroold:/data/scratch continues to be transferred to the new scratch file system /data/scratch/OLD

Please note: The synchronization is repeated on a daily basis. Changes in /data/scratch/OLD might get overwritten by subsequent syncs.

Phase III: shut down

starts: 2015-06-03

closed: 2015-09-03

• login old: The old allegro login node is shut down
• data: The file content in allegroold:/home and allegroold:/data/scratch is transferred one last time to the new scratch file system /data/scratch/OLD
• old data: The file content in allegroold:/home and allegroold:/data/scratch is deleted.

-- Cluster.salzmann - 29 Apr 2015