When updating low-level software for large computer systems it is di cult to verify whether performance requirements are met or not. Common practice is to measure the performance only when the new software is fully developed and has reached system veri cation. Since this gives long lead-times it becomes costly to remedy performance problems. Our contribution is that we have deployed a new method to synthesise production workload. We have, using this method, created a multi-core cache characteristics model. We have validated our method by deploying it in a production system as a case study. The result shows that the method is su ciently accurate to detect changes and mimic cache characteristics and performance, and thus giving early characteristics feedback to engineers.We have also applied the model to a real software update detecting changes in performance characteristics similar to the real system.