惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

IT之家
IT之家
博客园 - 聂微东
雷峰网
雷峰网
Microsoft Azure Blog
Microsoft Azure Blog
WordPress大学
WordPress大学
Hugging Face - Blog
Hugging Face - Blog
S
SegmentFault 最新的问题
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
T
Tailwind CSS Blog
博客园 - 三生石上(FineUI控件)
V
Visual Studio Blog
博客园 - 司徒正美
爱范儿
爱范儿
月光博客
月光博客
阮一峰的网络日志
阮一峰的网络日志
博客园_首页
博客园 - 【当耐特】
Jina AI
Jina AI
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
酷 壳 – CoolShell
酷 壳 – CoolShell
大猫的无限游戏
大猫的无限游戏
小众软件
小众软件
人人都是产品经理
人人都是产品经理
V
V2EX

博客园 - Clingingboy

冒个泡,刷刷存在感 使用文件映射和信号量来进程间通信 Xperf Basics: Recording a Trace (the easy way)(转) Xperf Basics: Recording a Trace(转) Android相关sdk使用 Uniscribe文字自动换行 Chrome RenderText分析(2) c++智能指针 codepage IMLangCodePages GUI 快捷键的实现思路 Menu实现逻辑 控件保持多种绘图状态的做法 2个函数宏技巧 绘图 Painter转接口封装的方式 DirectUI消息循环的简单封装 c++以代理的方式来实现接口化编程 c++对象工厂 使用模板来解决接口继承问题 VC++ 使用attributes定义接口
Xperf Analysis Basics(转)
Clingingboy · 2014-01-14 · via 博客园 - Clingingboy

2014-01-14 18:24  Clingingboy  阅读(582)  评论()    收藏  举报

FQ不易,转载

http://randomascii.wordpress.com/2011/08/23/xperf-analysis-basics/

I started writing a description of how to examine CPU Sampling data in xperf and realized that I needed a basics article to link back to. Here it is. Learn this before you go further.

Symbols

Without symbols you can’t do much, so configuring your symbol path is important. I use a variation on this batch file prior to running xperfview. Configure the “localbuilds” and “OurSymbolServer” directories appropriately, but leave the Microsoft symbol server as is.

@rem Add local build directories
set SYMS=c:\build\localbuilds
@rem Put symbol servers at the end to avoid network trips looking for local symbols.
set SYMS=%SYMS%;SRV*c:\symbols*\\OurSymbolServer\symbols
@rem Microsoft’s symbol server.
set SYMS=%SYMS%;SRV*c:\symbols*http://msdl.microsoft.com/download/symbols
set _NT_SYMBOL_PATH=%SYMS%

If this dialog keeps popping up when you load symbols in xperfview:

image

then it means you’ve hit a UAC (User Account Control – admin/non-admin) bug. Follow the link to solve it.

You can also configure symbols by going to Trace->Configure Symbol Paths.

You load symbols by going to Trace->Load Symbols, or right clicking and selecting Load Symbols.

It’s worth mentioning that symbols are compressed and cached to c:\symcache. This is an efficient symbol cache that is used exclusively by xperfview. Set the _NT_SYMCACHE_PATH variable if you want to configure the location. I just leave it at its default.

Zoomin’ In, Zoomin’ Out

Navigating the main window in xperfview is mostly a matter of zooming in and out. The obvious ways of doing this are in the right-click context menu. This menu lets you zoom in to the current selection, or zoom all the way out (Unzoom). I find this set of options frustrating. I like to zoom in and out a little bit. Zoom to selection is okay, I guess, but to have the only zoom-out option be “all the way out” is a bit silly.

The solution is to make sure you have a mouse with a mouse wheel. Hold down the control key and roll the wheel. That’s the way real programmers do it. I can’t use xperfview without a mouse wheel.

Enabling/Disabling/Reordering the Graphs

Xperfview defaults to having a particular set of graphs visible. The choice of graphs is biased towards people who are working on optimizing Windows itself, and is not as good a choice for those of us who are not actively working on the Windows scheduler. With that in mind, I recommend that you click on the flyout on the left side of the xperfview window and enable and disable graphs until it matches the correct set of graphs, conveniently shown here:

image

If you want CPU sampling by CPU, then go for it, but I think you’ll find it is usually not worth it.

In addition, you can change the order of graphs in order to get all of your favorites together. Just drag them around in this list to reorder them, and then double-click on the one you want to see.

Xperfview will remember the enabled/disabled state. That is good.

Xperfview will not remember the order of the graphs. You can’t win them all.

Selections

Selecting a region of interest in a trace is important because that lets you select which part of the trace you will invoke a summary table for. The diamonds on the CPU Scheduling graph and some other graphs are useful because they mark key points and because when selecting you can snap to them.

Once you’ve got a selection you can replicate it to other graphs by right clicking and selecting Clone Selection. This is often very useful.

That’s all folks

That’s enough for now. I’ll add more basics later if they seem worth it.