From Macro to Micro in C++

on YouTube ↗06m

" summary

Real data shows repeated microsecond spikes traced to container resizing rather than vague "load" issues, backed by added timers and a perf profile that surfaces a rehash call deep in templates. The surprise is the exact bucket sequence: prime bucket sizes, e.g. 389 → 769 rehash that matches the boost::multi_index growth policy, and the diagnosis is concrete — expensive allocations and element rehashing at specific sizes cause the outliers, not algorithmic complexity elsewhere.

AI-written summary · about

" chapters (7)

1. 0:00Intro
2. 0:22Meet Conor
3. 0:40Realtime metrics
4. 1:08Graphs
5. 2:27Prime Numbers
6. 3:17Bucket Sizes
7. 4:41Summary

" related talks

• 33mMemory Profiling for Mere MortalsThorsten Lorenz · JSConf Colombia 2016
• 01h 04mAdvanced .NET debugging techniques from real world investigationsKevin Gosse, Christophe Nasarre · NDC Oslo 2019
• 44mObservability Tools C++: Beyond GDB and printf - Tools to Understand the Behavior of Your ProgramIvica Bogosavljevic · CppCon 2022
• 16mMicroMon: A Monitoring Framework for Tackling Distributed HeterogeneityRamakrishnan Durairajan, Sudarsun Kannan, Baber Khalid, Nolan Rudolph · 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage '20)
• 30mUnderstanding the runtime behaviors of C++ programs using uftrace toolHonggyu Kim · CppCon 2017
• 26mProfiling Mastery: Decode Your Data and Elevate ObservabilityFrederic Branczyk · Observability Day North America 2024
• 32mIs Your Kernel Being Honest? Understanding & measuring low level bottlenecksFerris Ellis · Monitorama PDX 2024
• 25mUse counters to count thingsFred Moyer · Monitorama PDX 2024