如何找到程序的哪一部分花费最多时间（包括内核时间和上下文切换）
在对 CPU 密集型工作进行性能分析时，我通常使用 sudo perf record -a -s -e cpu-clock -g
我的具体案例： 我有一个从 gcs 加载数据并反序列化的程序。
我已经验证峰值吞吐量可以>200MBps，但是初始化时间需要4.5秒，其中1秒用于身份验证，另外3.5秒用于真正加载数据。
数据初始化读取约 300 个文件的元数据，但每个文件仅读取 1.25MB。
这是我的具体案例的抽象背景。我想知道linux(ubuntu)上是否有可用的工具，看看我的程序的哪一部分占用了最多的时间？这样我就可以对其进行一些更改。
When doing performance analysis on CPU-intensive work, I generally use sudo perf record -a -s -e cpu-clock -g <executable>. But that doesn't show how much time is spent on context-switch, how much time is spent in network stack, which I think is not super helpful to a network-IO-bounded task.
My specific case:
I have a program which load data from gcs, deserialize it.
I have verified the peak throughput could be >200MBps, but initialization time takes 4.5 seconds, of which 1 second is spent on authentication, the other 3.5 seconds is really loading the data.
The data initialization reads metadata for ~300 files, but each file only read 1.25MB.
This is the abstract context for my specific case. I want to know if there's any tools available in linux(ubuntu), to see which part of my program takes the most of the time? So that I could make some changes to it.
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论