Abstract

Many applications benefit from repeatability: being able to run the same program deterministically on the same input, hence seeing the same output, across both time (reproducing an earlier run) and space (reproducing the same execution on a different machine, say). Two major use cases are bug reproduction in everyday software, and distribution of scientific software. In this very informal talklet I will sketch a case for supporting repeatability somewhere near the core of the operating system, rather than (as at present) in user-space tools (e.g. rr, UndoDB, CDE ) or hypervisors (revirt, VMware Workstation). Time permitting, I’ll speculate wildly about the potential benefits to storage subsystems, toolchain design, software build and deployment, compiler bootstrapping, debugging, and the optimisation of recurring time/space trade-offs.