The distributed data management system Rucio manages all data of the ATLAS collaboration across the grid. Automation, such as data replication and data rebalancing are important to ensure proper operation and execution of the scientific workflow. In this proceedings, a new data allocation grid service based on machine learning is proposed. This learning agent takes subsets of the global datasets and proposes a better allocation based on the imposed cost metric, such as waiting time in the workflow. As a service, it can be modularized and can run independently of the existing rebalancing and replication mechanisms. Furthermore, it collects data from other services and learns better allocation while running in the background. Apart from the user selecting datasets, other data services may consult this meta-heuristic service for improved data placement. Network and storage utilization is also taken into account.

Top