2022-05075 - Post-Doctoral Research Visit F/M Data Allocation Strategies for Energy Optimization in the Qarnot Computing Cloud Infrastructure.
Le descriptif de l’offre ci-dessous est en Anglais

Type de contrat : CDD

Niveau de diplôme exigé : Thèse ou équivalent

Fonction : Post-Doctorant

Niveau d'expérience souhaité : Jeune diplômé

Contexte et atouts du poste

Within the framework of a partnership (you can choose between)

  • Collaboration between Inria and Qarnot Computing

The postdoc will mainly take place in the Topal/Storm team (supervision Olivier Beaumont (Topal), Laercio Lima Pilla (Storm) and Lionel Eyraud-Dubois (Topal)) of the Inria center at the University of Bordeaux, with occasional stays at Qarnot Computing (frequency and duration to be discussed with the candidate, Qarnot Computing and Inria). The objective is to propose an efficient solution for data placement under energy constraints and to publish the research work in the best journals and conferences (on Energy and Parallel and distributed algorithms and infrastructures). 


Mission confiée

The joint challenge between Inria and Qarnot aims to develop and promote best practices in geo-distributed hardware and software infrastructures for intensive computing with a reduced environmental footprint.


As long as datacenters are not located closer to cities, the reuse of server heat will essentially require the decentralization of datacenters. As such, the distribution of the storage system, as an essential building block of the computing platform, is a prerequisite for the reuse of the waste heat emitted by the servers.


The distribution of the storage system raises many challenges to which this project will provide answers. In this context, we are interested in data that are likely to be used many times ( training,...). In order to allow a good availability of the computations, it is necessary that the data be replicated on several sites, so as to maximize the probability of finding a free resource on a site which has the data. Otherwise, it will be necessary to move the data on demand, which consumes energy and introduces delays. The replication itself introduces an energy overhead, due to the cost of the initial transfer and the resource occupation.


The question to be solved is therefore the following: given

- a set of data

- a set of tasks associated with this data (or a distribution of the probability that a task needing a specific data appears in the system)

- a cost model (time and energy) of storage, transfers (in emergency or during off-peak periods) 

- a quality of service guarantee for users and 

- a distribution of the probability of availability of resources on the different sites, 

find a data replication policy that guarantees the quality of service while minimizing the energy cost.


The introduction of all the constraints and the multi-objective optimization problem is too complex to be addressed directly and we will use an incremental approach allowing the gradual integration of the constraints and objectives, which may also involve restricting the type of applications considered. Another difficulty concerns the collection of data (on the behavior of applications, on the modeling of storage and transfer costs, etc.). The objective is to build an on-line placement strategy (for which the data is known as it is received) that can also make dynamic redistribution decisions.


Principales activités

Main activities (5 maximum) : Algorithm Design, Modeling, Simulation




* bibliographical study, and in parallel analysis of usage data provided by Qarnot (Qarnot, Inria)

* design of allocation algorithms in a simple model (offline, possibly homogeneous) (Inria)

* (optional) development of a simulation environment to validate different approaches (Qarnot, Inria)

* design of an online algorithm, and validation in simulation (Qarnot, Inria)




Technical skills and level required : distributed algorithms, infrastructure modeling

Languages : possible options are possible at this stage

Relational skills : ability to exchange with designers of a complex full-scale solution and to extract the most relevant parameters for modeling, allowing the design of relevant algorithms. 



  • Subsidized meals
  • Partial reimbursement of public transport costs
  • Possibility of teleworking and flexible organization of working hours
  • Professional equipment available (videoconferencing, loan of computer equipment, etc.)
  • Social, cultural and sports events and activities
  • Access to vocational training
  • Social security coverage


2653€ / month (before taxs)