This thesis studies the performance impact of the Two Phase Commit
mechanism used in order to achieve consistency in a distributed PostgreSQL
system. This impact is compared to the performance gained
by removing Two Phase Commit and abandoning consistency. The
tests are run on different number of nodes in the system to find a link
between the performance hit and the size of the system. Performance
is measured using the unit Queries per Second.
Firstly, a short introduction is given to demonstrate why and when
distributed database systems is necessary and the difficulties it poses.
Thereafter follows a description of the system constructed to conduct
the experiment, together with the results of this experiment. The thesis
is concluded with a discussion regarding the results and thoughts
on future variations of the experiment.
The conclusion of the experiment is that Two Phase Commit puts
a low maximum performance on the system, especially when compared
to the alternative of abandoning consistency. The number of
nodes did not significantly affect the performance of the system using
Two Phase Commit, however the alternative had a clear negative
correlation between numbers of nodes and Queries per Second.