L'Agenda du Libre

Quatrième événement du Printemps de l'innovation Open Source, organisé par le GTLL de Systematic et l'Irill, présidé par Roberto Di Cosmo

Open Source pour le Cloud et les Conteneurs

Programme dirigé par Frédéric Lepied (Red Hat) et Gilles Muller (Inria / LIP6)

• 10:00 - Everything you want to know about virtualization on large multicores

  Vivien Quéma (IMAG), Gaël Thomas (Telecom Sud-Paris), Pierre Sens (LIP6), Gauthier Voron (LIP6)

  Cloud providers are now proposing large virtual machines with several tens of cores. Since such virtual machines often relies on complex hardware architectures, new challenges arise for hypervisors to achieve good performances. This talk will present some of these problems and why they happen.

• 10:45 - Reclaiming memory to the right containers

  Julien Sopena (LIP6)

• 11:30 - Towards microservices architecture to transcode videos in the large at low costs

  Olivier Barais et David Bromberg (IRISA).

  The increasing popularity of videos over Internet, combined with the wide heterogeneity of various kinds of end users’ devices, imposes strong requirements on the underlying infrastructure and computing resources to meet the users expectations. In particular, designing an adequate transcoding workflow in the cloud to stream videos at large scale is: (i) costly, and (ii) complex. Lightweight containers technologies such as docker are more and more adopted for deploying and managing micro-services architectures [1]. Indeed, they hide a part of the complexity resulting from the language heterogeneity and the large choice of existing implementation frameworks that can be used to implement modern applications. Containers provide those benefits without introducing the general cost of the virtualization.

  It exists lots of open-source configuration managers such as kubernetes[2], nomad[3], apache Mesos[4], or docker swarm[5] to define and keep a standard modular applications in a consistent state and to dispatch its services on a cluster of remote nodes. However, there is still a lack of adopted open-source solution for easily designing and distributing a dataflow (i.e. a complex directed-acyclic-graph of tasks for processing data) on a distributed infrastructure to improve the performance of such applications. It exists some initial solution such as TaskCluster[6] from Mozilla dedicated to Continuous integration but it still remains the needs to design these data-flows and to efficiently schedule the execution of its dependent tasks.

  Through this presentation/demo, we are giving our experience feedbacks of building both a low cost and efficient transcoding platform over an ad hoc computing cloud built around a cluster of Raspberry Pis and containers solutions that use docker, hypriot / cluster labs [7] and go-docker [8]. We want in particular discuss the need to provide better abstractions for easily managing hardware capabilities reservation such as GPU from the workflow.

• 14:00 - Traefik, un reverse-proxy/load-balancer dynamique écrit en GO

  Emile Vauge (Containous)

• 14:45 - Piloter le réseau de vos conteneurs dans OpenStack avec Kuryr

  Christophe Sauthier (Objectif Libre)

• 15:30 - Skydive: debugging and monitoring your networks from containers to the physical packets

  Sylvain Afchain (Red Hat)

• 16:15 - PAUSE
• 17:00 - How the OpenShift PaaS is implemented on top of Docker/Kubernetes

  Chmouel Boudjnah (Red Hat)

• 17:45 - Enjeux de l’orchestration des microservices

  Hervé Leclerc (Alter Way)

• 19:00 - Conclusion et buffet-networking

Inscription gratuite mais obligatoire

Autres journées à venir :

L'Agenda du Libre | Frama-C Day - Analyse et vérification de code L'Agenda du Libre | Techniques de programmation web à l'état de l'art