Dynamic Linkers Are the Narrow Waist of Operating Systems

While software applications, programming languages, and hardware have changed, operating systems have not. Widely-used commodity operating systems are still modeled after the ones designed in the seventies. The accumulated burden of backward compatibility with the large software ecosystems that run our workloads prevents systems from embracing more efficient and disruptive designs explored by the system research community. This paper advocates a fresh approach to operating system research, where innovations are incrementally integrated into operating systems, without disrupting existing software, to gradually reshape our daily-use systems. The dynamic linker emerges as a pivotal element in this transformation process, redefining system behavior. The paper outlines specific use cases, covering performance enhancements, strengthened security measures, streamlined software deployment, and enriched programming language abstractions. Additionally, the paper introduces Spidl, an experimental modular dynamic linker to facilitate the exploration of this promising new research avenue.