The Production and Analysis system (PanDA system) has continuously been evolving in order to cope with rapidly changing computing infrastructure and paradigm. The system is required to be more dynamic and proactive to integrate emerging workflows such as data carousel and active learning, in contrast to conventional HEP workflows such as Monte-Carlo simulation and data reprocessing.
Intelligent Data Delivery Service (iDDS) is an experiment agnostic service to orchestrate workload management and data management systems, in order to transform and deliver data and let clients consume data in near real-time. iDDS has been actively developed by ATLAS and IRIS-HEP. iDDS has a modular structure to separate core functions and workflow-specific plugins to meet a diversity of requirements in various workflows, simplify the development and operation of new workflows, and provide a uniform monitoring view. The goal of iDDS is the seamless integration of new workflows as well as to address performance issues and suboptimal resource usage in existing workflows.
This talk will report architecture overview of iDDS, orchestration of PanDA and Rucio for optimal storage usage in data carousel, dynamic task chaining in ATLAS production system with instant decision making for active learning, data streaming with on-demand marshaling to minimize data delivery from data ocean to analysis facilities and users, integration of iDDS with other workload management systems, and plans for the future.