Programme

Keynote 1: “Unikernels: Library Operating Systems for the Cloud -- Contributions for Compatibility, Security and Heterogeneity Support”

Résumé: Unikernels are minimal, single-purpose virtual machines. This new type of operating system can be seen as the evolution of the Exokernel/LibOS model applied to the cloud. Unikernel bring promises of numerous benefits within many application domains in terms of lightweightness, performance, and inter-application security. However, the unikernels model is not without flaws. In this talk, I will describe a set of contributions aiming at solving fundamental problems regarding unikernels. A first major issue is the difficulty/impossibility to port existing application to most modern unikernels. I will present a solution in the form of a system capable of running as a unikernel a binary application originally compiled for a popular traditional OS, Linux. A second issue I will cover regards the security concerns caused by the lack of intra-unikernel isolation. I will present a proposed solution that is the introduction of such isolation in a lightweight fashion using a modern intra-address space compartmentalization mechanism, Intel Memory Protection Key. Finally, I will present a system leveraging unikernels to allow the migration of native compute-intensive/HPC applications between x86-64 servers and ARM64 embedded systems for improving job throughput in the data center.

Session 1 : 80 min, Chair : Thomas Ropars

• A Deep Reinforcement Learning approach to Network Slice Placement in multi-domain networks. José Jurandir Alves Esteves Junior, Salah Bin Ruba, Amina Boubendir, Fabrice Guillemin
• A Dynamic Task Scheduler Tolerant to Multiple Hibernations in Cloud Environments. Luan Teylo Gouveia Lima, Luciana Arantes, Pierre Sens, Lúcia Drummond
• Optimisation du placement des licences logicielles dans le Cloud pour un déploiement économique et efficient. Arthur CHEVALIER, Eddy CARON
• Amélioration et adaptation de méthodes d'apprentissage par renforcement neuronal pour le placement optimal des réseaux virtuels. Maxime Elkael, Hind CASTEL-TALEB, Badii Jouaber

Session 2 : 60 min, Chair : Boris Teabe

• Balancing your service load with model predictive control. Adrien WION, Mathieu Bouet, Luigi Iannone, Vania CONAN
• Big data applications scheduling on ephemeral and heterogeneous Cloud resources. Handaoui Mohamed
• BMC : Proxy et cache noyau sûrs pour l'accéléreration de Memcached. Yoann Ghigoff
  

Keynote 2: “Les défis d'une offre de Cloud  Provider On premise” 

Résumé: L'engouement autour du Cloud Computing a poussé plusieurs administrateurs d'infrastructure informatique à proposer une offre de Cloud à leur entreprise. La définition d'une telle offre de service d'effectue progressivement, ceci au rythme du renouvellement et de l'urbanisation du nouveau matériel informatique. Cette migration engendre divers problèmes, les uns liés à la mise en place et l'exploitabilité d'une infrastructure hétérogène en intégrant le paradigme Cloud. Les autres sont liés à l'adoption de cette nouvelle offre de service pour les utilisateurs finaux. Au cours de cette présentation, nous mettons en exergue les challenges d'une offre de cloud privé. Nous présenterons également les difficultés de mise en place des bonnes pratiques adoptées et mises en oeuvre par les provider publics ou celles issues des travaux effectués par les experts académiques.

Session 3 : 80 min, Chair : Alain Tchana

• Dynamic Service Placement and Load Distribution in Edge Computing. Adyson Magalhaes Maia, Yacine Ghamri-Doudane, Dario Vieira, Miguel Franklin de Castro
• Everyone should benefit from Hardware Assisted Virtualization, not only the Hypervisor. Thévenon Gaspard, Stella Bitchebe, Alain Tchana, Vo Quoc Bao Bui
• FaaSCache: Système de cache mémoire opportuniste et sans surcoûts pour le FaaS. Kevin Nguetchouang, Lucien Ngale, Stephane Pouget, Djob Mvondo, Mathieu Bacou, Renaud Lachaize
• LSTM-based radiography for anomaly detection in softwarized infrastructures. Alessio Diamanti, stefano secci, José Manuel Sanchez Vilchez

Session 4 : 80 min, Chair : Mathieu Bacou

• The emergence of NVRAM: An opportunity for resilient and energy-efficient Cloud applications. Thomas Ropars
• Cloud Allocation and Consolidation Based on a Scalability Metric. Tarek Menouer, Amina Khedimi, Christophe Cérin
• Derniers développements dans SimGrid. Martin Quinson
• Dynamic composition of services in the Edge. Marie Delavergne

Keynote 3: “DPDK – the next performance leaps for new hardware processing units”

Résumé: From 2000 to 2010, innovations of packet processing were based on Network Processors. From 2010 to 2020, the General-Purpose CPUs became the new Network Processors thanks to the new software packet processing stacks. In a cloud environment, it means that these CPUs need a trade-off between running the legacy applications and having dedicated CPU cores for packet processing. Thanks to the latest innovations of many PCIe-based smart devices, it became wise to rethink packet processing using the DPUs (SmartNICs) in order to better use the GPUs and the NVMe devices. This solution relieves many cycles that remain available for the core CPUs in order to increase the density of applications into the cloud environments. Thomas Monjalon (DPDK lead maintainer) and Vincent Jardin (DPDK co-founder) will provide an analysis of the upcoming challenges for cloud and high-performance networking combining DPDK with the new PCIe devices.

Session 5 : 80 min, Chair : Djob Mvondo

• The FogGuru Living Lab: A LoRa-enabled fog computing testbed. Guillaume Pierre
• MOSAICO : Multi-layer Orchestration for Secured and low lAtency applICatiOns. Mathieu Bertrand
• On-Node Correlation and Prediction Based Data Reduction in WSN for Smart Agriculture. Christian Salim, Nathalie Mitton
• Adaptive network device services in a virtualized environment. Brice EKANE

Session 6 : 80 min, Chair : Redha Gouicem

• Optimization of resource allocation in software-defined wide area network SD-WAN. Ons fares, Abdulhalim Dandoush, Nadjib Ait-Saadi, Ilhem Fajjari
• Ordonnancement de conteneurs. Tarek Menouer
• Resource Allocation for Multi-Tenant Edge Computing. Andrea Araldo
• Scalevisor: Un hyperviseur distribué pour le rack. Yohan Pipereau, Gaël Thomas

Keynote 4: “Advanced traffic engineering for data center networks”

Résumé: In the last decade, the ever increasing volume of data collected and processed by applications has boosted the adoption of cloud and distributed computing. Big Data applications such as MapReduce, Spark, Giraph, Pregel are deployed in large clusters of machines that may be distributed over different geographical locations. All the traffic issued by these applications must be managed efficiently so has to improve the utilization of the infrastructure and the completion times of data-intensive processing tasks. In this talk, we will present ongoing work on the acceleration of “multi-stage data flows” where computation phases are interleaved with communication phases. The set of flows in one communication phase is called a coflow, or a shuffle phase in MapReduce. Then, we will present scalable load balancing and rate control algorithms that operate only from edge switches with distributed agents. These algorithms aim at minimizing the maximum link utilization or maximizing the network utility, and they provide any time feasibility and diminishing returns towards near-optimal solutions. Finally, we will quickly highlight other contributions on elephant flow scheduling and routing with guaranteed latency.

Session 7 : 80 min, Chair : Thomas Begin

• Scheduling Service Function Chains with Dependencies in the Cloud. Tarek Menouer, Amina Khedimi, Christophe Cérin, Mohammed CHAHBAR
• Buffer overflow. Yves KONE, Stella Bitchebe, Alain Tchana, Pierre Olivier
• Travaux en NFV/Cloud au LISTIC. Salamatian Kavé, Sébastien Monnet
• Mitigating vulnerability windows with hypervisor transplant. Ngoc Tu Dinh, Boris Teabe, Alain Tchana, Gilles Muller, Daniel Hagimont