Fig. 1

Conceptual view of DRMaestro’s architecture. In a loop-based approach, the Resource manager (e.g., Kubernetes [10]) appends to the queue the new incoming jobs. If there are enough resources in the cluster, a list of jobs is sent to the Flow Network Model module to update the models. The optimal placement is determined by sending the updated models to the Flow Network Solver Model (e.g., Firmament [13]). If necessary, DRMaestro starts the disaggregated resource daemons