一种基于eBPF的并行程序在线性能数据采集方法
By using eBPF technology to collect performance data of parallel programs in real time, the problem of existing tools being unable to analyze data online is solved, enabling low-cost and efficient performance data acquisition and display, and improving the monitoring efficiency of parallel programs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2022-08-12
- Publication Date
- 2026-07-17
AI Technical Summary
Existing parallel program monitoring tools are mainly based on post-event analysis, which cannot achieve online performance data acquisition and display without modifying the source code, making performance problem localization difficult and costly.
A performance data acquisition module was developed using eBPF technology. It intercepts parallel library functions through the eBPF compiler bcc and the user dynamic probe uprobe to collect and transmit performance data in real time. The module also uses a stream processing module to match and identify inefficient operations, and finally displays the data online through a visualization module.
It enables real-time acquisition of parallel program performance data and identification of inefficient operations without restarting the parallel program, thereby reducing the cost and complexity of performance analysis and improving analysis efficiency.
Smart Images
Figure CN115328731B_ABST