Sync job between G5K and Fit (and XXX )
Let's focus on G5K + Fit
Goal
For an experiment that spans G5K + FIT, two jobs are required: one on G5K, one on FIT. As an experimenter I want the job to start (and end) at the same time.
Implementation hints
-
A naive algorithm would be to test different timesteps and actually try to reserve the resources on both platforms and rollback if the job get too desynchronized. An keep retrying until we get jobs synchronized. This is simple but with the risk of putting some extra load on the different schedulers (and reserving resources for nothing)
-
Another approach is to look at the reservation planning and try to get a good estimate of a common date to start.
Refs:
- G5K status:https://api.grid5000.fr/doc/3.0/#tag/status/paths/
13.01sites~1{siteId}~1status/get - Fit drawgantt: https://api.iot-lab.info/#/drawgantt/getDrawganttNodes
- Chameleon/Blazar: https://docs.openstack.org/api-ref/reservation/v1/index.html?expanded=list-allocations-detail,get-allocation-detail#resource-allocations