Investigation of Methods and Tools to Profile a Virtual Platform

Abstract: In the embedded system development industry, virtual platforms are used heavily for emulating the target hardware to shorten the testing phase and obtain a faster time-to-market for the product. To make the virtual platforms better performing in respect of time, an increased observability of the system is necessary. Tools and techniques for increased visibility exist but their uses have not been documented for virtual platforms yet. This thesis examines the existing tools available and use them in a case study of Ericsson’s SVP startup time. A simplified version of the SVP has been used to perform the profiling and tracing. The parameters - modules, registers and sockets - are analyzed as they determine the size of the virtual platform. These parameters are used while profiling the platform to find a correlation between the startup time and their respective quantity. Using the profiling and tracing tools on the studied virtual platform, it is possible to determine the correlation between the startup time and the parameters. From the results obtained, the cost of creation of one unit of module, register and socket can be determined. For a large size virtual platform with an important amount of communication between the modules, the results obtained from the case study show it is beneficial to use the IP-XACT register with a static top. Additionally, the profiling and tracing help to explain delays such as processes waiting to be killed. These results can be utilized to make virtual platforms more time efficient and to identify and troubleshoot bottlenecks.