Program

Program

Wednesday 9

8h30-9h15     Welcome coffee

9h15 – 9h30    Opening

9h30-10Hh30 Keynote 1: Michel Mandjes

                           Chair: Bruno Tuffin (Inria)

10h30-11h     Coffee break

11h-12h30  Session Scheduling, queuing systems & Resource allocation (3 talks)

                       Chair: Corinne Touati (Inria)

12h30-14h      Lunch

14h – 15h30  Session Modeling & Performance (3 talks)

                          Chair: Balakrishna Prabhu (CNRS)                        

 15h-30-16h   Break

16h – 17h       Keynote 2: Nadia Oudjane

                         Chair: Yezekael Hayel (Avignon Université)

Welcome drink


Thursday 10

9h-10h           Keynote 3: Sergio Grammatico

                        Chair: Tijani Chahed (Telecom SudParis)

10h – 10h30   Break

10h30 – 12h30 Session Pricing & economic models (4 talks)

                             Chair: Alonso Silva (Nokia Bell Labs)

14h – 15h30     Invited session

                           Chair: Ana Busic (Inria & ENS Paris)

  • George Kesidis
    • Title: Analysis of a System with Multiple Caches that Share Objects
  • Nicolas Gast
    • Title: Asymptotic Optimality in Restless Bandit
    • Abstract: Restless bandit and weakly coupled MDPs are classical modeling tools to model resource allocation problems (for instance in scheduling or queuing networks). These problems are computationaly hard to solve and there has been a number of heuristics proposed in the literature, based on linear relaxation and mean field control. While some of them date from the 90s (the famous Whittle index), there has been a surge of recent development in this area. This talk will discuss these later developments: the what, the how and the why.
  • Panagiotis Andrianesis
    • Title: Distributed Economic Dispatch in Power Networks Harnessing Data Center Flexibility

    • Abstract: In this work, we focus on the interaction of Data Centers (DCs) with electric power networks in an economic dispatch problem, harnessing the ability of DCs to serve as flexible loads that can alter their power consumption to alleviate network congestion. Since DCs can transfer workload between each other, controlling their output via some demand response mechanism that respects power generation and network constraints, while also accounting for DC Quality of Service (QoS) can achieve system-wide benefits. From the DC perspective, we aim to explore the benefits of their incorporation in an economic dispatch problem and demonstrate that they can achieve significant cost savings for the entire system (data and power networks), enabling the IT sector to contribute to societal sustainability efforts.

      Our main contributions are as follows. First, we leverage results from queuing theory to model DCs and form QoS-based cost functions – signifying how well a DC can carry out its workload given an amount of active servers, for which we provide convexity guarantees under certain conditions. Second, we integrate DCs in a centralized economic dispatch problem that determines, apart from power generation, DC workload shifting and server utilization, while respecting power network constraints. Third, we provide a tractable decentralized formulation of the economic dispatch problem employing Lagrangian decomposition and a primal-dual algorithm, which can cater for both the power network constraints, and, most importantly, the DC workload shifting, in a distributed manner that scales for the “coupled” data and power networks. Fourth, we present experimental results on a standard power network that provide useful insights on the system-wide benefits from harnessing DC flexibility in the economic dispatch problem, emphasizing on the trade-offs between the DC locations, their efficiencies, and QoS costs.

      Keywords: Data centers, flexibility, economic dispatch, Lagrangian decomposition.

      This is joint work with Athanasios Tsiligkaridis, Ayse Coskun, Michael Caramanis, and Ioannis Paschalidis (Boston University). Part of this work is supported by France 2030 under grants ANR-22-PETA-0004 and ANR-22-PETA-0009 in the frameworks of the AI-NRGY and FlexTASE projects.


Best paper Award

Gala dinner

Friday 11

9h-10h          Keynote 4: Michael Jordan

                        Chair: Hélène Le Cadre (Inria)

10h-10h30     Break

10h30-12h     Session Energy, Generative AI (3 talks)

                          Chair: Patrick Maillé (IMT Atlantique)

12h15-12h30  Closing