Skip to Main content Skip to Navigation

Analyse et ordonnancement d’un système hiérarchique virtualisé composé d’applications temps réel strictes

Abstract : In the context of smart cities, the concept of autonomous vehicle is gaining importance in the objective of carrying out surveillance missions for the protection of infrastructures. In this context, unmanned aerial vehicles (drones) embed several applications (control command, communication and video) of different criticality with often heterogeneous associated certification constraints. The instrumentation and control application with the highest level of criticality must be certified at the highest level of requirement. When multiple applications of different criticality run on the same hardware platform, the most critical application contagiously imposes certification on the less critical applications to the same level of requirement as that of the most critical application. This is not always possible given the complexity of the applications and can lead to a certification cost which can be significant. One solution to this problem is to use a virtualization system based on a certified hypervisor, providing spatial and temporal isolation of applications. Each application then runs in a dedicated virtual machine (VM). The activation of VMs is managed by the hypervisor in charge of this isolation.From a scheduling point of view, this forms a two-level hierarchical system. VMs are scheduled by the hypervisor at the first level, the tasks of each VM are scheduled at the second level by a specific scheduling algorithm. We consider in this thesis a fixed priority scheduling at the second level.Our work fits into the context of the CEOS surveillance drone project. In this project, the goal is to create an autonomous surveillance drone for non-military purposes. This is a FUI project with various partners such as INRIA, Enedis, EDF, Sysgo, Alerion, Thalès, Caen airport. This drone is defined for three use cases: (i) monitoring power lines in the EDF network, (ii) monitoring water pipes in the mountains for ENEDIS and (iii) monitoring fences in the Caen airport.For the CEOS project, a type 1 hypervisor developed by the Sysgo company (PikeOs) was chosen. PikeOs is certified DAL-A for various avionics standards such as ARINC 653 and MILS. It is thanks to its certifications that this hypervisor was chosen. One of its peculiarities is that it is possible to master the level 1 scheduling of VMs. For level 1, the scheduling of virtual machines using a scheduling table must be determined beforehand, no algorithm is imposed. Regarding the tasks executed by the VMs, their scheduling is determined by the real-time operating system associated with the virtual machine. It is also possible to introduce tasks directly into PikeOS which will then be scheduled by a fixed priority scheduling.The freedom to choose an algorithm for first-level scheduling involves several questions: How to schedule VMs so that all tasks in their VM meet their deadlines? Under the conditions of predetermined scheduling for virtual machines, which scheduling would give a good success ratio for the scheduling of tasks ?
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Thursday, January 27, 2022 - 4:00:10 PM
Last modification on : Sunday, February 13, 2022 - 9:21:20 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03545962, version 1



Tristan Fautrel. Analyse et ordonnancement d’un système hiérarchique virtualisé composé d’applications temps réel strictes. Intelligence artificielle [cs.AI]. Université Gustave Eiffel, 2021. Français. ⟨NNT : 2021UEFL2011⟩. ⟨tel-03545962⟩



Record views


Files downloads