Converged, Efficient and Safe Architecture based on Near Memory Accelerators
— Preview
Rethink hardware so that it is co-designed with the application, with a view to converged architecture and trust in an environment with abundant data to process
Denis Dutoit, Head of the computing and digital architectures program at CEA-List
Keywords: Digital architecture, rack-scale computing, disaggregated heterogeneous architectures, memory-centric architectures, hardware gas pedals, energy efficiency, design, formal methods
European cloud sovereignty includes control over processors and gas pedals. With Dennard’s Law coming to an end and Moore’s Law slowing down, processor performance will evolve towards more parallelism (multi-core), more specialization (gas pedals), and integrating new interconnections. Virtualization, by separating hardware resources from logical ones, facilitates converged architectures, making the cloud more modular, fast and agile. However, this increased complexity makes the intermediate adaptation layer harder to validate and more prone to failure.
The Archi-CESAM project proposes to rethink hardware (computation, memory and interconnection) so that it is co-designed with the application in a converged and trusted architecture perspective, in an environment known for its abundance of data to be processed. The Archi-CESAM project tackles this major Cloud evolution with a global, coordinated approach to distributed architectures, acceleration, interconnection and security bricks, not forgetting design methods.
— Missions
— Our researches
— Partners
Consortium
Inria, CEA, Université de Rennes, Télécom SudParis, GINP, CNRS
— Research team
9 permanent staff (not funded by the project)
19 people funded by the project, including :
9 PhD students
4 post-docs
6 engineers
Our teams in France
Other projects
CARECloud
Understanding, improving and reducing the environmental impact of cloud computing
Archi-CESAM
Efficient and safe converged architecture based on near-memory gas pedals / Archi-CESAM
DiVa
Virtualized disaggregation
Taranis
Model, deploy, orchestrate and optimize cloud applications and infrastructures
PEPR Cloud projects
Spirec
Multi-level supervision and prediction for geo-distributed and heterogeneous infrastructures in the Cloud/Edge/IoT continuum
Steel
Efficient and secure data storage and processing on cloud-based infrastructures
Silecs
Super Infrastructure for Large-Scale Experimental Computer Science for Cloud/Edge/IoT
TrustinCloudS
Cloud infrastructure security
