Application-specific unikernels have the reputation for being complicated to create. Instead, being binary compatible with Linux promises that you can use a unikernel as a simple drop-in replacement, especially for running container workloads. But how can we avoid redeveloping Linux for compatibility reasons? What about fork, clone and co. while maintaining a single address space? Where can we still specialize and keep performance benefits?

In this talk we will dive into our efforts over the last year with Unikraft for binary compatibility with Linux. We'll discuss the challenges we have solved, the lessons we have learned, and the optimizations we have been able to make. In all of this, we have set the goal of not requiring to recompile any application code.