P2IO has brought support to the development and management of common computer infrastructures: shared access computer rooms/data centers, innovative distributed file storage and backup software solutions. On the physics application side, support was also given to advanced architecture and algorithmic R&D to exploit parallelism for data selection and analysis tasks. In the next phase, P2IO will foster and support emerging efforts in the labs to solve the upcoming computing challenges of data processing and simulation in our scientific community:

• The intensive computing challenge: processing and simulating large physics datasets, ranging from complex detector events to telescope images, require massive computing resources, and advanced architectures. Notably, astrophysics simulation will benefit from the upcoming exa-scale supercomputing infrastructures. Efficient parallelism and efficient architecture are two key aspects to match rapidly evolving processors with the physics workflow in an automatic and sustainable way, to produce software stacks and data pipelines intended to live over several decades.
• The big data challenge: on the I/O side, the recent explosion of data volume recorded or simulated by experiments introduces new data storage and locality challenges. New tools inspired by "big data" approaches need to develop. On the algorithmic side, Artificial Intelligence approaches have proven to be very efficient for big data problem solving, and could play an increasing role in our experimental and theoretical physics paradigms. Deep machine learning methods are a promising avenue in this prospect, the foundation of many R&D efforts in our community.

P2IO is involved in this research, benefiting from the close links with data science community, in particular the UPSaclay Center for Data Science, an interdisciplinary network of data scientists.