2024-08-04, 10:30–11:00 (Asia/Taipei), TR411
Hermes, a versatile time-series profiling tool, offers comprehensive system performance analysis. It operates by executing periodic or oneshot jobs as defined in the configurations, thereby monitoring performance metrics. To minimize impact on system performance, Hermes intelligently skips metric collection if the system loading remains below defined thresholds. This feature ensures that the profiling service can seamlessly run in production environments without causing any disruptions. At present, Hermes supports various approaches, including perf, ftrace, and ebpf, to gather detailed information for performance analysis. This diverse range of supported methods enhances the tool's effectiveness in providing insightful performance metrics. During this presentation, Chengen will delve into the concept and implementation with meticulous detail, providing a comprehensive overview of the subject matter.
Hermes 是一種時序效能分析工具的實踐，透過定期或一次性執行預先定義的任務來提供全面的系統效能分析。為了最大程度地減低對系統效能的影響，若系統負載處於該任務預定的閾值以下，Hermes 將跳過此次任務的效能指標收集。此功能可確保效能分析服務即便於生產環境中也能無縫運行，不會造成任何服務中斷。目前 Hermes 支持多種方法來收集效能指標，包括 perf、ftrace 和 ebpf，藉此增強工具提供深入效能指標的可能性。在本次演講中，Chengen 將深入探討此工具的概念和實作。
目前任職於 Ubuntu，過往工作有 Virtualization / Distributed systems / Linux kernel 等開發經驗。對系統效能分析相關技術具熱誠，目前致力於累積分析經驗及更深入的理解 Linux kernel。