„Broadcast Yourself“

Abstract

Reproducing bugs is hard. Deterministic replay systems address thisproblem by providing a high-fidelity replica of an original programrun that can be repeatedly executed to zero-in on bugs. Unfortunately,existing replay systems for multiprocessor programs fall short. Thesesystems either incur high overheads, rely on non-standard multiprocessorhardware, or fail to reliably reproduce executions. Their primarystumbling block is data races -- a source of nondeterminism thatmust be captured if executions are to be faithfully reproduced.In this paper, we present ODR--a software-only replay system thatreproduces bugs and provides low-overhead multiprocessor recording.The key observation behind ODR is that, for debugging purposes, areplay system does not need to generate a high-fidelity replica ofthe original execution. Instead, it suffices to produce any executionthat exhibits the same outputs as the original. Guided by this observation,ODR relaxes its fidelity guarantees to avoid the problem of reproducingdata-races altogether. The result is a system that replays real multiprocessorapplications, such as Apache, MySQL, and the Java Virtual Machine,and provides low record-mode overhead.