Adaptively optimising virtual machines (VMs) have found widespread usage, which unite many advantages, including good performance and platform independence. Alas, due to high implementation efforts only a few VMs can deliver excellent performance. Recent research has focused on language implementation frameworks, which promise good trade-offs between performance and implementation efforts, including the idea of meta-tracing. However, meta-traced VMs currently have bad warmup characteristics – meaning that the time taken to reach peak performance is high. In my research I want to develop and evaluate novel techniques to tackle the issue of slow warmup.