In academia, analyzing various performance aspects of software is nowadays quite well-researched. Theoretically, one can automatically analyze the complexity of functions, generate time-consuming inputs, or efficiently profile the software with minimal overhead. However, while the results are always exciting, applying these techniques in practice is not so straightforward.

For over seven years, we have developed Perun: a performance management system and a tool suite. We usually evaluated our techniques on smaller or medium-sized projects (at most half a million lines of code). So, in the recent year, we decided to move towards a bigger challenge: the Linux kernel.

In this talk, we will summarize our experience and the challenges we have faced when we moved from the academic field into the real world.