Moving Compute Contingents Between HPC Systems (e.g. to Otus)

The principal investigator (PI) of a compute project can request to redistribute the granted compute time between our HPC systems by mail to pc2-support@uni-paderborn.de. This can be of particular interest when new systems become available that didn’t exist at the time of proposal writing. As we recently inaugurated the Otus system, the migration process is currently available for all project types. Any remaining contingent on Noctua 1 get automatically migrated. There are two modes to re-distribute compute contingents between HPC systems.

A Relative Distribution (recommended)

You specify to distribute the granted compute resources between our HPC systems in percentages. The change only affects the remaining runtime of the compute project.
Example: You have a compute project with a runtime of 12 months and a granted total of 24 Mio. CPU-core-hours on Noctua 2. Nine months into the project, you want to modify the distribution to Noctua 2 and Otus, let’s say 50%/50%. For the last three months, you have (24 Mio. CPU-Core-h total contingent)/(12 months project runtime)*(3 months of remaining runtime)=6 Mio. CPU-Core-h contingent that can be distributed. With the 50%/50% distribution Noctua 2 and Otus each get 3 Mio. CPU-Core-h contingent for the three months. This effectively results in a split of the 30-day contingent from the previous 2 Mio. CPU-Core-h (24 Mio. CPU-Core-h/12 months) to the two systems, resulting in 30-day contingents of 1 Mio. CPU-Core-h for each system.
The situation is summarized in the following table:

An Absolute Distribution

Alternatively, you can specify to move an absolute portion of your granted compute contingent between systems. However, you should be aware of the boundary condition that originates from the continuous-usage-concept: You can only move a portion of your contingent that is still available. Importantly, the available contingent here is NOT the difference between granted and used compute time but the granted compute contingent relative to the remaining runtime.
Thus, you can move at most (total granted compute contingent)/(total project runtime)*(remaining project runtime) which is the contingent in the remaining project runtime.

Example: You have a compute project with a runtime of 12 months and a granted total of 24 Mio. CPU-core-hours on Noctua 2. Nine months into the project, you want to move some compute resources from Noctua 2 to Otus.
You can move at most (24 Mio. CPU-core-hours total granted)/(12 months total project runtime)*(3 months remaining runtime)=6 Mio. CPU-core-hours for the remaining 3 months to Otus, even if you have more unused resources overall.