Summary

In recent years, Grids have been used extensively for a wide range of scientific purposes making this infrastructure gradually a more accepted alternative to conventional cluster computing. However, even today the reliability of computational sites in Grids tends to leave a lot to be desired. We argue that a priori estimation of site reliability will give users the opportunity to decrease the likelihood of job failure, thereby reducing excess consumption of grid resources and improving the user experience of utilizing the Grid. This work focuses on an in-depth analysis of user-perceived reliability on grid infrastructures and proposes a usercentric approach to reduce the number of failures on grids. We proceed by proposing a platform-independent design for a site reliability prediction tool and a description for a platform-specific implementation. In conclusion, we provide validation experiments to give an impression of the potential added value of using a reliability prediction tool. 3