The objective of this thesis is the analysis of a fundamental problem of virtualization, namely the examination of the impact of dependability insufficiencies occurring in a runtime platform onto the operation of applications. The significance of the problem originates in the widespread use of virtualized environments without specific quality guarantees for mission-critical applications. The evolution of the hardly needed category of carrier grade virtualized platforms necessitates as first step deep knowledge about fault impacts onto critical services.
The aim of the thesis is to examine the sensitivity of faults of virtualization technologies used in large system by means of the software-based fault injection (SWIFI) approach, a frequently used approach in the simulation based accelerated assessment faults that occur in real world. It examine the influence of artificially injected faults onto the platform and through this the impact of applications.
The first chapter defines the problems and presents an overview of virtualization techniques.
One of the essential tool of dependability evaluation is fault injection. Traditional error log analysis needs a long period of observation time for collecting sufficient data for statistically correct results. Fault injection campaigns based on artificially injection of faults into the system simulates faulty behavior in an accelerated way, thus it is capable to produce enough data within a relatively limited period. In order to gain useful results with respect to the general principles of metrology, measurements that reflecting the reality should be carried out, thus the faithfulness of experiments necessitates a faithful fault set.
Fault injection into the large systems is one of the most common methods of empirical dependability assessment, as impact analysis of an appropriate fault ensures the coverage of the most frequent fault events and conditions. This method requires minimal knowledge about the system and can be performed on almost any system; however it differs from an ideal measurement due to the complexity of the problem and to the measuring principle.
The second and the third section details the principles of dependability evaluation and fault injection.
The purpose of the thesis is to design an architecture that tries to assure a conformance to the general principles of metrology in order to facilitate deep, but general conclusions by means of a detailed analysis of the data.
The demonstration of the architecture and the measurements is presented in the fourth, fifth and sixth sections. This last section presents the functioning of the system by examples.
The seventh section evaluates the results and suggests topics for further development.