Lightning Talk: Adventures in Benchmarking Timestamp Taking in Cpp
summary
Benchmarks compare clock_gettime vs rdtsc and use QuickBench then real servers; initial QuickBench numbers show 85× slower for clock_gettime, but running compiled code on target hardware shows only ~2× slowdown. Profiling with perf reveals clock_gettime ultimately uses the CPU TSC on Linux so execution stays in user space, with extra wrapper instructions accounting for the difference. Net result: timestamps cost about 15 ns more for the wrapper, not a syscall-level catastrophe.
AI-written summary · about
related talks
01h 38mKeynote: Benchmarking - It's About Time
05mLightning Talk: Micro C++ Benchmarks & Murphy's Law
01h 11m“Benchmarking C++ Code"
31mBenchmarking and Optimization of Rust Libraries
59mCounting Nanoseconds Microbenchmarking C++ Code
01h 00mMeasuring Qt Performance
47mBenchmarking C++, From video games to algorithmic trading
01h 29mWhen Nanoseconds Matter: Ultrafast Trading Systems in C++