Publicação
Evaluation of hypervisors for embedded and real-time systems
| Resumo: | Virtualization has shown immense promise within embedded systems, mostly due to its ability to manage and optimize the allocation of the finite resources typically present on these systems and therefore contributing to the system’s overall cost reduction while maintaining a low execution overhead, considering the time constraints of each subsystem as well as providing isolation between different VMs. This dissertation focuses on providing a benchmark for hypervisor testing using embedded systems as the underlying hardware, providing a collection of modules for system testing, as well as tools for result analysis and fault injection implementation. Regarding result analysis, a new fault categorization model was proposed, the CRIH scale, with the intent of providing users with a deeper understanding regarding fault effects in the tested software. Besides the benchmark implementation, a case study regarding hypervisor performance and fault tolerance is also documented, serving as proof of the benchmark’s application. Using an embedded system as the underlying hardware, a comparison of several hypervisors was performed, resulting in an in-depth comparison of two mature hypervisors: KVM and Xen. In the performed experiments, KVM has shown to have more performance across the board when compared to Xen, completing executions in less time overall. However, Xen possesses less result dispersion, indicating that it would be more suitable for handling realtime scenarios, where having less result deviation ensures that fewer deadlines will be missed. Finally, this dissertation’s contributions aim to allow system manufacturers to ship out their systems faster, providing them with a platform featuring modules that are representative of real autonomous driving features, as well as providing them with more analysis tools regarding hypervisor tradeoffs, promoting safety in terms of system design. |
|---|---|
| Autores principais: | Teixeira, Rui Pedro de Castro Brito |
| Assunto: | Hypervisor Virtualização Confiabilidade Sistemas Embebidos Teses de mestrado - 2020 |
| Ano: | 2020 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | Virtualization has shown immense promise within embedded systems, mostly due to its ability to manage and optimize the allocation of the finite resources typically present on these systems and therefore contributing to the system’s overall cost reduction while maintaining a low execution overhead, considering the time constraints of each subsystem as well as providing isolation between different VMs. This dissertation focuses on providing a benchmark for hypervisor testing using embedded systems as the underlying hardware, providing a collection of modules for system testing, as well as tools for result analysis and fault injection implementation. Regarding result analysis, a new fault categorization model was proposed, the CRIH scale, with the intent of providing users with a deeper understanding regarding fault effects in the tested software. Besides the benchmark implementation, a case study regarding hypervisor performance and fault tolerance is also documented, serving as proof of the benchmark’s application. Using an embedded system as the underlying hardware, a comparison of several hypervisors was performed, resulting in an in-depth comparison of two mature hypervisors: KVM and Xen. In the performed experiments, KVM has shown to have more performance across the board when compared to Xen, completing executions in less time overall. However, Xen possesses less result dispersion, indicating that it would be more suitable for handling realtime scenarios, where having less result deviation ensures that fewer deadlines will be missed. Finally, this dissertation’s contributions aim to allow system manufacturers to ship out their systems faster, providing them with a platform featuring modules that are representative of real autonomous driving features, as well as providing them with more analysis tools regarding hypervisor tradeoffs, promoting safety in terms of system design. |
|---|