xperf问题 [英] xperf questions

问题描述

我是使用xperf的新手，我有几个问题：

1。我使用xperf同时在NT内核记录器上打开PROFILE和CSWITCH提供程序。我注意到图表"cpu sampling by cpu"之间存在显着差异。和"cpu使用cpu"，我可以假设这个差异的原因是什么，所以我使用后一个图表是
，因为在我看来它更准确。 问题是当我做汇总表时我无法按模块聚合使用情况，因为没有模块列，而不是采样的图形cpu使用情况。我错过了什么吗？是否有
可以在第二个图表中查看模块的使用情况？

2。查看图表中的摘要表 "按流程划分的CPU使用率"我假设从CSWITCH而不是PROFILE获取数据，我不知道每个进程花费多少时间在每个CPU上（在多核CPU上）  - 没有CPU列。
如何检索此信息。

3。在计算CPU利用率时，不同的使用率图是否考虑了cpu缩放（当时钟频率动态变化时）？

谢谢

解决方案

1）

我对这个问题的解释是以下假设：

 

如果不是这种情况，我会使用符号功能并按线程检查CPU采样。   如果不是这种情况，我会使用符号功能并按线程检查CPU
采样。

这使您可以详细检查和分析cpu重量。

并且在所有情况下都使用"隐形" taskmanager上的cpu活动 - >我通过帖子找到了他们。

 

关于Q2）

我有同样的愿望过去。我转储我的etl并导出信息。很酷的是我得到的结果是csv。我只是复制粘贴使用过的cpu内核信息并使用自动摘要功能:)


我在第9频道的WPT产品经理现场演示的精彩视频中找到了这个技巧。 / p>

Q3）我很害怕，我不确定你对此的意思：（。

Hi,

I am a novice in using xperf and I have several questions:

1. I use xperf with both PROFILE and CSWITCH providers open on NT kernel logger. I notice there is a significant difference between the graphs "cpu sampling by cpu" and "cpu usage by cpu", I can assume what is the reason for that difference, so I am using the latter graph since it seems to me more accurate. The problem is that when I do summary table I can't aggregate the usage by module since there is no module column as opposed to the graph cpu usage by sampling. Am I missing something ? Is there a way to see the usage by module in the second graph ?

2. When looking at the summary table from graph  "CPU usage by process" which I assume gets its data from CSWITCH and not from PROFILE, I can't tell how much time each process spent on each CPU (on a multi core CPU)  - there is no CPU column. How can I retrieve this information.

3. Do the different usage graphs takes into account the cpu scaling (when the clock frequency is dynamically changing) when they calculate CPU utilization ?

Thanks

解决方案

Hi,

for 1)

My expirions around this issue is the following assumption:

 

If this is not the case I would use the Symbols function and check the CPU sampling by thread.   If this is not the case I would use the Symbols function and check the CPU sampling by thread.

And this allow you to check and analyze in detail the cpu weight.
And in all cases with "invisible" cpu activity at the taskmanager --> i found them by thread.

 

About Q2)

I had the same wish in the past. And I dump my etl and export the information. The cool thing was that I get the result as csv. And I just copy paste the used cpu kernel information and use the auto summarize function :)
I found this trick at the nice video from the WPT product manager live demo at Channel 9.

Q3) I'm afraid, I'm not sure what you mean with this :(.

这篇关于xperf问题的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！