Building CPU stubs to optimize CPU bound systems: An application of dynamic performance stubs.

Dynamic performance stubs provide a framework for the simulation of the performance behavior of software modules and functions. Hence, they can be used as an exten- sion to software performance engineering methodologies. The methodology of dynamic performance stubs can be used for a gain oriented performance improvement. It is also possible to identify “hidden” bottlenecks and to prioritize optimization possibilities. Nowadays, the processing power of CPUs is mainly increased by adding more cores to the architecture. To have benefits from this, new software is mostly designed for parallel processing, especially, in large software projects. As software performance optimizations can be difficult in these environments, new methodologies have to be defined. This paper evaluates a possibility to simulate the functional behavior of software algorithms by the use of the simulated software functionality. These can be used by the dynamic performance stub framework, e.g., to build a CPU stub, to replace the algorithm. Thus, it describes a methodology as well as an implementation and evaluates both in an industrial case study. Moreover, it presents an extension to the CPU stubs by applying these stubs to simulate multi-threaded applications. The extension is evaluated by a case study as well. We show show that the functionality of software algorithms can be replaced by software simulation functions. This stubbing approach can be used to create dynamic performance stubs, such as CPU stubs. Additionally, we show that the concept of CPU stubs can be applied to multi-threaded applications.