Assessment of Virtual Machine Working-Sets in SimuBoost
- Type:Bachelor Thesis
- Date:12.03.2018
- Supervisor:
Prof. Dr. Frank Bellosa
Marc Rittinghaus - Graduand:Johannes Werner
- Links:PDF
-
Abstract:
Distributing checkpoints over a network and restoring them in an FSS can take up to several minutes. This is partly due to the fact that in SimuBoost’s current implementation the whole RAM image has to be loaded into the FSS.
In this thesis, we assess whether restoring only the working-set can lead to a decrease in checkpoint loading times. We do this by using an FSS to measure the working-set of different workloads and different interval lengths. The results show that even though the size of the working-set depends heavily on the workload and the interval length, it is still an only small portion of the whole RAM itself with the biggest measured working-set only being 7.94%. We also assessed an optimization technique similar to post copy. Instead of the whole working-set, only the write working-set is restored initially and the remaining pages have to be loaded on demand.
Bibtex:
@mastersthesis{werner18workingset,
author = {Johannes Werner},
title = {Assessment of Virtual Machine Working-Sets in SimuBoost},
type = {Bachelor Thesis},
year = 2018,
month = mar # "12",
school = {Operating Systems Group, Karlsruhe Institute of Technology (KIT), Germany}
}