Search

Over the last decades, several data preservation efforts have been undertaken by the HEP community, as experiments are not repeatable and consequently their data considered unique. ARCHIVER is a European Commission (EC) co-funded Horizon 2020 pre-commercial procurement project procuring R&D combining multiple ICT technologies including data-intensive scalability, network, service interoperability and business models, in a hybrid cloud environment. The results will provide the European Open Science Cloud (EOSC) with archival and preservation services covering the full research lifecycle. The services are co-designed in partnership with four research organisations (CERN, DESY, EMBL-EBI and PIC/IFAE) deploying use cases from Astrophysics, HEP, Life Sciences and Photon-Neutron Sciences creating an innovation ecosystem for specialist data archiving and preservation companies willing to introduce new services capable of supporting the expanding needs of research. The HEP use cases being deployed include the CERN Opendata portal, preserving a second copy of the completed BaBar experiment and the CERN Digital Memory digitising CERN’s multimedia archive of the 20th century. In parallel, ARCHIVER has established an Early Adopter programme whereby additional use cases can be incorporated at each of the project phases thereby expanding services to multiple research domains and countries.

With the increasing number of Machine and Deep Learning applications in High Energy Physics, easy access to dedicated infrastructure represents a requirement for fast and efficient R&D. This work explores different types of cloud services to train a Generative Adversarial Network (GAN) in a parallel environment, using Tensorflow data parallel strategy. More specifically, we parallelize the training process on multiple GPUs and Google Tensor Processing Units (TPU) and we compare two algorithms: the TensorFlow built-in logic and a custom loop, optimised to have higher control of the elements assigned to each GPU worker or TPU core. The quality of the generated data is compared to Monte Carlo simulation. Linear speed-up of the training process is obtained, while retaining most of the performance in terms of physics results. Additionally, we benchmark the aforementioned approaches, at scale, over multiple GPU nodes, deploying the training process on different public cloud providers, seeking for overall efficiency and cost-effectiveness. The combination of data science, cloud deployment options and associated economics allows to burst out heterogeneously, exploring the full potential of cloud-based services. With the increasing number of Machine and Deep Learning applications in High Energy Physics, easy access to dedicated infrastructure represents a requirement for fast and efficient R&D. This work explores different types of cloud services to train a Generative Adversarial Network (GAN) in a parallel environment, using Tensorflow data parallel strategy. More specifically, we parallelize the training process on multiple GPUs and Google Tensor Processing Units (TPU) and we compare two algorithms: the TensorFlow built-in logic and a custom loop, optimised to have higher control of the elements assigned to each GPU worker or TPU core. The quality of the generated data is compared to Monte Carlo simulation. Linear speed-up of the training process is obtained, while retaining most of the performance in terms of physics results. Additionally, we benchmark the aforementioned approaches, at scale, over multiple GPU nodes, deploying the training process on different public cloud providers, seeking for overall efficiency and cost-effectiveness. The combination of data science, cloud deployment options and associated economics allows to burst out heterogeneously, exploring the full potential of cloud-based services.