Analyzing and improving GNOME startup time

The startup time of open source desktop applications and environments does not compare well with other systems, especially closed-source systems. We perform a detailed analysis of the startup of the GNOME desktop environment using a mixture of available open-source tools and ad-hoc measuring techniques, identifying bottlenecks and examining strategies to improve performance. Our results show that startup is I/O bound and dominated by disk seeks, and that substantial improvements can be made at relatively little cost. We validate the results of the analysis with proof-of-concept code modifications which provide a 40% reduction in measured login time; many of these suggestions have been adopted by the developers and implemented in the current releases of the applications in question. We also evaluate the impact of library loading strategies to startup time, finding that trivial changes to the dynamic linker can provide a 10% reduction in the startup time of large applications.