Impact of (nested) virtualization on benchmark accuracy

This project has already been solved by one student at Faculty of Informatics (Masaryk University) and its continuation will be the subject of his Master Thesis. However, as the project leader is looking for alternative solutions, this project is best suited for students from other Faculties (FIT BUT, FBE Mendelu), so the topic is not being solved twice at the same Faculty.

The side-effects of virtualization to benchmarks have been known for a long time. However, virtualization technologies are advancing, and their footprint is smaller and smaller. In this thesis, the student should investigate how individual virtualization technologies: full (KVM, vbox, oVirt) or shared kernel (Docker, chroot, mock) or nested and mixed, are affecting performance and stability of various Java benchmark types – CPU, time, network, bytecode, IO. Support for virtualization is seen also in JDK itself, so more than one Java can come to play. We will provide the student with benchmarks, pros and cons of this setup, VMs and images and other virtualization know-how, hardware and several JDKs and their internals and setups. The student will be able to focus on the running itself, and deep analysis of results and improvement of the topic itself.


Project Resources