Software/Hardware And unKnown EnviRonment Interactions (SHAKER) team

The team develops methods and tools for optimizing software and hardware systems which factor in the constraints and hazards associated with their environment.

Methods

(Joint) modeling of hardware and software architectures
Design of dynamically adaptable Sw/Hw systems depending on the environment and application requirements
Online verification of system behavior and temporal properties
Implementation of tools for design, simulation and tracing.

Targets

embedded system,
system-on-chip,
network of sensors,
edge/cloud infrastructure.

Areas

autonomous vehicles,
critical systems,
IoT,
industry of the future,
space,
marine environment,
home support

Collaborations

Companies: ATOS, Orange, Naval Group, Thalès and DDN among others
Institutions: CEA, b<>com Institute of Research & Technology (IRT), NIST (US) and Taiwan Academy of Sciences among others
Laboratories: INRIA, IETR, LIRMM and LAMIH among others
CNRS Research Groups: RSD, SOC2

Examples of research projects

Energy optimization of data storage systems for high-performance computing (HPC) applications

Objective of this project: to develop methods and tools for modeling data access profiles in an efficient and non-intrusive manner then use the models established to develop strategies for optimizing the energy consumption of compute nodes during the data access stage.
Financed by: Atos
Led by: Jalil Boukhobza, research professor at ENSTA Bretagne/Lab-STICC (SHARP department, SHAKER team)

In performance terms, storage systems represent one of the most significant weak links in an IT system, particularly for applications which process large amounts of data such as in the field of high-performance computing (HPC).

The emergence of new storage technologies is an opportunity to reduce the performance gap between working memory and storage, as well as energy consumption.
These technologies are deployed at various levels:

- storage memory (e.g. 3DxPoint),
- its interface (e.g. NVMe),
- or its software management (e.g. object store).

These technologies imply significant growth in the complexity of storage management in order to meet the service quality requirements of applications.

Intrusion detection and Security/Performance/Energy compromise; study for fleets of drones (DISPEED)

Objective: outline a model and strategies for running intrusion detection systems embedded on drones equipped with heterogeneous architectures and striking a relevant compromise according to the attack criticality and state of the system and mission, between detection speed/energy consumption.

Financed by: AID
Partners: FORTH Greece, Naval Group, UBO
Led by: Jalil Boukhobza, research professor at ENSTA Bretagne/Lab-STICC (SHARP department, SHAKER team)

The distributed operation of fleets of drones during missions makes them vulnerable to diverse attacks that it is crucial to detect. Embedded in these drones are hardware components (computing and storage) with heterogenous computing power and energy consumption for performing the various tasks necessary for their mission.

The project sets out to develop models, strategies and tools for optimizing the energy cost of intrusion detection on a fleet of drones or any other cooperative system with major energy or hardware capacity constraints. These systems operate in cooperation to accomplish a joint mission. The network load therefore varies enormously depending on the context of the mission, which means that the intrusion detection system does not need to be run continuously on equipment requiring significant hardware capacity or consuming a considerable share of the system’s energy.

The aim of the project is thus to study and analyze how the performance of the IDS can be adapted using various hardware components depending on this network load and the context of the mission.

4 challenges underpin the project:

Challenge 1: modeling the execution environment
Challenge 2: setting up an assessment platform
Challenge 3: designing a configuration selection strategy based on multi-objective optimization tooling
Challenge 4: implementing an inter-drone balancing/offset computing system for reducing energy consumption.

HPC data positioning on multi-tiered and heterogeneous storage systems (DATAMESS)

Objective: design effective data positioning systems on multi-tiered storage architectures in the field of high-performance computing.

Financed by: CEA
Led by: Jalil Boukhobza, research professor at ENSTA Bretagne/Lab-STICC (SHARP department, SHAKER team)

In performance terms, storage systems represent one of the most significant weak links in an IT system, particularly for applications which process large amounts of data. The emergence of new storage technologies is an opportunity to reduce the performance gap between working memory and storage. These technologies, deployed at the level of storage memory (e.g. 3DxPoint), its interface (e.g. NVMe) and even its software management (e.g. object store), imply significant growth in the complexity of storage management. In addition, amid the "big Data" boom, more and more applications are processing huge amounts of data, and present different levels of criticality.

Against this backdrop, we intend to study and come up with new data positioning strategies with different levels of criticality on heterogenous, geo-distributed storage systems. As part of this project, we will explore several techniques including machine learning, reinforcement learning and optimization methods, to guarantee effective online data positioning.