Уровень 0 · материалов: 3
В кластер входят документы, описывающие методы, инструменты и технические подходы к поиску и устранению проблем с производительностью программного обеспечения.
Общие признаки: выявление узких мест производительности, инструменты профилирования, оптимизация работы программного обеспечения, анализ производительности в Linux
Группа выше: Метрики производительности и их интерпретация
Смысл: The text serves as a comprehensive guide on how to identify performance bottlenecks in Go applications. It demonstrates the use of internal Go profiling tools (pprof, trace) and external Linux utilities (perf, SystemTap) to optimize CPU and memory usage.
A technical guide demonstrating how to use Go's internal pprof and trace tools alongside Linux's perf and SystemTap to optimize CPU and memory performance.
Смысл: The text announces the open-sourcing of Perforator, a high-performance profiling tool by Yandex that solves the 'broken stack' problem in Linux profiling using eBPF. Its main goal is to provide a scalable, low-overhead way for developers to identify performance bottlenecks in production environments across large clusters.
Yandex has open-sourced Perforator, an eBPF-based continuous profiling system for Linux that provides accurate stack unwinding with minimal overhead for large-scale distributed systems.
Смысл: The main idea is that third-party APIs can introduce severe performance bottlenecks due to inefficient internal implementations, and these can be identified and potentially bypassed using profiling, reverse engineering, and binary hooking techniques.
A developer uses profiling and reverse engineering to discover that the Skype COM API's inefficient use of system locale functions was slowing down their backup utility.