Close

Presentation

Birds of a Feather
:
HPC, Containers and Big Data Analytics – How can Cloud Computing contribute to the New Challenges
Event Type
Birds of a Feather
Passes
Tags
Big Data Analytics
Clouds and Distributed Computing
Containerized HPC
HPC workflows
Storage Technologies
TimeTuesday, June 26th10am - 11am
LocationSubstanz 1, 2
DescriptionCloud computing architectures provide a large diversity of system configurations, from virtual servers to dedicated hosts and bare metal, all managed through a consistent interface. Several providers have also started to include HPC-type configurations with Infiniband and MPI library options. Furthermore, data management tools enable massive parallel data processing and analytics, with the support of containers, when applicable. Future systems are expected to be based on hybrid models, with static workloads being supported by dedicated resources and dynamic workloads being supported by external providers to gather the full economic benefits of the pay-per-use model. This would also offer new and better options for sharing data (access, communication and synchronization), that will have to be reconsidered on all available parallelization levels. Moreover, it imposes new challenges regarding automation of dynamic workflows across hybrid infrastructures.
HNSciCloud is a European pre-commercial procurement (PCP) project co-funded by the European Commission Horizon 2020 Work Programme, kicked-off in January 2016. The PCP is developing a set of cloud services addressing the needs of a group of buyers from across Europe that serve multiple data-intensive research communities, including the needs for HPC workloads. Partners in the project led by CERN are CNRS, DESY, EMBL, ESRF, IFAE, INFN, KIT, STFC, EGI, SurfSARA and Trust-IT. The goal is to establish a hybrid platform using open source solutions, integrating the resources in data centres operated by the buyers group with those of European cloud players, to create a competitive marketplace of services for a wider range of users.