Understanding the Performance-Effectiveness Trade-Offs of AddressSanitizer in Real-World Programs: An Empirical Study

Abstract

AddressSanitizer (ASan) is widely deployed in industrial C/C++ development, but its runtime and memory costs can limit routine use in continuous integration and production-like testing. Although many optimized ASan-like sanitizers have been proposed, heterogeneous workloads, configurations, and baselines make their practical trade-offs difficult to compare. We present a unified, deployment-oriented empirical study of ASan and seven representative optimized sanitizers: ASan--, SanRazor, GiantSan, RSan, MEDs, ASAP, and LFP. Using 17 SPEC CPU2006 benchmarks, Juliet, MSET, 21 real-world CVEs, and a multi-threaded MySQL case study, we examine component costs, parameter sensitivity, build configurations, detection outcomes, and deployability. In our ordered cumulative ablation, the largest first-step runtime reduction follows removal of memory-access checks, while the largest late-stage RSS reduction occurs at the heap-management step. Parameter tuning identifies lower-memory configurations without additional false negatives relative to default ASan on Juliet, although this result does not establish preservation of delayed-UAF detection. The reported evidence shows distinct efficiency, detection, and deployment outcomes, but unequal completion sets prevent a formal cross-tool ranking. Our results support treating tuned ASan as a necessary, configuration-dependent comparison baseline for sanitizer evaluation and provide bounded evidence for deployment decisions under different resource and risk constraints.