Timing side channels---an infamous class of security vulnerabilities 
resulting in leakage of confidential information through differences 
in the running time of a system---constitute an active research topic 
since the mid 1990's.  Mitigation techniques for such side channels 
include program transformations intended to remove any dependence of a 
program's execution time on its confidential inputs. However, results 
by Coppens, Verbauwhede, Bosschere, and Sutter [Coppens2009] show that 
some of the proposed program transformations do not account for 
microarchitectural influences and therefore leave a program vulnerable 
to timing attacks. This thesis investigates three selected program 
transformations presented in the literature [Agat2000, Molnar2005, 
Barthe2006] in the context of programs executed on a Java virtual 
machine. By the help of information theoretic results of 
Chatzikokolakis, Chothia, and Guha [Chatzikokolakis2009], we are able 
to quantitatively assess the effect of each program transformation on 
information leakage.

In particular, we developed a sample-based analysis framework in order 
to characterize the timing behavior of a given Java program and 
quantitatively estimate the induced information leakage. We present 
four case studies including non-trivial Java programs with side 
channels which are transformed by the selected program transformations 
and analyzed by our framework. We evaluate the effects of each program 
transformation with respect to mitigation of information flow and 
induced execution time overhead. Additionally, we manually replace 
certain program fragments in a step-wise manner, allowing conclusions 
about which fragments contribute to a timing channel.

Our results include an empirical confirmation of the inappropriateness 
of two of the three selected program transformations for Java virtual 
machines with just-in-time compilers. Furthermore, we identify timing 
channels caused by recursion and other general program elements.