Seminars

Friday 29 May, 10am - 1pm Service Optimization and Orchestration across the Computing Continuum ESILV, L307

• Function Placement for In-network Federated Learning

  Nour El-Houda Yellas (Télécom SudParis / SAMOVAR) [description]
  Federated learning (FL), particularly when data is distributed across multiple clients, helps reducing the learning time by avoiding training on a massive pile-up of data. Nonetheless, low computation capacities or poor network conditions can worsen the convergence time, therefore decreasing accuracy and learning performance. We propose a framework to deploy FL clients in a network, while compensating end-to-end time variation due to heterogeneous network setting. We present a new distributed learning control scheme, named In-network Federated Learning Control (IFLC), to support the operations of distributed federated learning functions in geographically distributed networks, and designed to mitigate the stragglers with lower deployment costs. IFLC adapts the allocation of distributed hardware accelerators to modulate the importance of local training latency in the end-to-end delay of federated learning applications, considering both deterministic and stochastic delay scenarios.

  Fermer

  [slides]

• Faster Latency Constrained Service Placement in Edge Computing with Deep Reinforcement Learning

  Emmanuel Hyon (Université Paris Nanterre / LIP6) [description]
  This presentation presents a novel approach to address an edge computing optimization problem by modeling it as a combinatorial optimization problem combining multi-commodity flow and linear latencies constraints. We then develop an equivalent linear formulation of the Service Placement Problem, allowing us to use traditional Integer Linear Programming (ILP) methods. Facing the problem complexity, we develop a use-case-based heuristic that provides a first practical placement solution. We also propose a Reinforcement Learning (RL) methodology to model the network configuration under orchestration actions.

  Fermer

  [slides]

• From Cloud to Constellation: Orchestrating Services at the Edge of Space

  Farah Aït Salaht (DVRC / ESILV) [description]
  The rise of LEO satellite constellations is extending the Cloud–Edge continuum into space, turning satellites into compute nodes capable of hosting services and processing data directly in orbit. This paradigm, often called Orbital Cloud, raises new challenges in scheduling, resource allocation, service placement and decision making, amplified by satellite mobility, the intermittency of visibility windows, and limited on-board resources. In this talk, we will introduce the problem and its main stakes, provide a state of the art of modeling and optimization approaches from the recent literature, identify open issues that remain underexplored, and outline promising research directions for this emerging field.

  Fermer

  [slides]

Monday 8 December, 3pm - 5pm Devoteam X ESILV - Cloud Frugal & Autonome ESILV, Salle du Conseil

• ACADS: A Framework for Adaptive Cost-Aware Deployment of Stream Processing System in the Cloud

  Farah Ait-Salah, Laetitia Della Maestra, Daniel Wladdimiro (ESILV) [description]
  The increasing volume of data from Internet of Things (IoT) devices, digital services, and user interactions has made cloud-based Stream Processing Systems (SPS) essential for real-time analytics. A primary challenge in deploying SPS in the cloud is efficiently provisioning resources: both the replication degree of each operator and the selection of virtual machines (VMs)—in type and quantity—must be carefully determined to balance performance and cost. In this work, we introduce ACADS (Adaptive Cost-Aware Deployment and Scaling), a unified planning model that formulates this provisioning task as a Constraint Satisfaction Problem (CSP). ACADS integrates logical and physical deployment decisions by jointly optimizing operator replication and VM placement, guided by workload characteristics and infrastructure constraints. Leveraging a queue-theoretic performance model and real-world pricing from …

  Fermer

• DaMoOp: A global approach for optimizing denormalized schemas through a multidimensional cost model

  Jihane Mali, Nicolas Travers (ESILV) [description]
  The complexity of database systems has increased alongside the exponential growth of data, necessitating Information Systems (IS) architects to continuously refine data models and meticulously select storage and management options that align with requirements. While existing solutions focus on data model transformation, none offer guidance in selecting the most suitable model. In this context, we propose DaMoOp, an automated approach for leading data model selection process. DaMoOp starts from a conceptual model and associated use case comprising queries, settings and infrastructure constraints, to generate relevant logical data models. A cost model, considering environmental, financial, and temporal factors, facilitates comparison and selection of the most suitable data model. Our cost model incorporates both data model and queries costs. Additionally, we suggest a data model selection process that …

  Fermer

• Autonomous Cloud

  Jeremy Mechouche (Devoteam) [description]
  Fermer

Monday 10 November, 3pm - 5pm Student Research Track ESILV, L519

• Cross-Layer Cloud Orchestration for Dynamic IaaS and FaaS Deployment

  Constantin Keuky (ESILV) [description]
  Fermer

• AI for Reactive IoT Service Deployment in Edge Networks with Energy Awareness

  Théo Hardy (ESILV) [description]
  Fermer

Wednesday 7 July, 10am - 12.30am Cloud Optimization ESILV, L408

• Elasticity and flexible stream processing

  Alessio Pagliari (LiP6) [description]
  Fermer

• A Framework for Adaptive Cost-Aware Deployment of Stream Processing System in the Cloud

  Farah Ait-Salah, Laetitia Della Maestra, Daniel Wladdimiro (ESILV) [description]
  The increasing volume of data from Internet of Things (IoT) devices, digital services, and user interactions has made cloud-based Stream Processing Systems (SPS) essential for real-time analytics. A primary challenge in deploying SPS in the cloud is efficiently provisioning resources: both the replication degree of each operator and the selection of virtual machines (VMs)—in type and quantity—must be carefully determined to balance performance and cost. In this work, we introduce ACADS (Adaptive Cost-Aware Deployment and Scaling), a unified planning model that formulates this provisioning task as a Constraint Satisfaction Problem (CSP). ACADS integrates logical and physical deployment decisions by jointly optimizing operator replication and VM placement, guided by workload characteristics and infrastructure constraints. Leveraging a queue-theoretic performance model and real-world pricing from …

  Fermer

• Liability and trust metrics

  Chrystel Gaber (Orange) [description]
  Fermer

• FACT-DM: A framework for automated cost-based data model transformation

  Jihane Mali, Nicolas Travers (ESILV) [description]
  As data continues to grow at an unprecedented rate, the complexity of database systems also increase significantly, requiring information systems (IS) architects to constantly adapt their data model and carefully select the optimal solution (s) for storing and managing data that align with new queries, settings, and constraints. Having a tool to visually show the impact of different changes, will help IS architects in their decision making. In this paper, we propose a framework to demonstrate the impact of denormalization of data models on their cost and consequently IS architects can choose the best trade-off.

  Fermer

View map location