Optimization of existing resource usage
Research conducted by the University of Bonn.
We have optimized the usage of our existing resources, and, as a result, maximized hardware potential, including opportunistic usage of desktop machines (Compute Site in a Box, CoSiaB) and backfilling HPC/HTC resources using COBalD/TARDIS.
Efficiency of HPC/HTC sites was increased with the help of:
- Backfilling:
- Desktop machines were integrated into backfilling by extending the existing queuing system.
- ClassAds and cgroups were adapted to reflect user activity and machine usage times.
- Creating a lightweight solution for different sites to deploy (Puppet)
- Monitoring backfilled jobs and impact on other usages.
Steps performed to achieve the results mentioned above:
- Backfilling our local HTC cluster since 2020
- Backfilling local HPC resources since 2020 using HTCondor overlay batch system
- Adpating HTCondor adstash to store monitoring data locally
- Ability of cluster to adapt quickly once users submit new jobs.
Future work
During FIDIUM extension, the following work is planned in the resource usage optimization area:
- Cooperation with other sites trying to deploy COBalD/TARDIS
- Finalization of connected older ideas on Compute Site in a Box
- Participation in AUDITOR to report provided resources (benchmarking all WN architectures).
New project devoted to renewable energy
FIDIUM members can collaborate to work on the new project. Its purpose is to consider developments in the renewable energy market, as we are obviously headed towards a 100 % Renewable Energy (RE) based power infrastructure. The two key paradigms to build upon are:
- There will never be a cheaper source of energy again than solar (and to a certain extent and for a limited time, wind).
- Volatility of source availability will (probably — with caveats depending on the cost & price of Power-to-X) mean that energy is for free when the sun is shining and the wind is blowing
The project will involve considering developments in the renewable energy market, and computing in the age of strong sector coupling.