vtune performance analysis tool--find out program performance bottlenecks

1 Overview

VTune is a relatively powerful performance analysis software from Intel. It mainly includes three gadgets:

(1) Performance Analyzer: Performance analysis to find the hot part of software performance, which is generally the key point of the performance bottleneck.

(2) Intel Threading Checker: used to find thread errors, can detect resource competition, thread deadlock and other issues

(3) Intel Threading Profiler: Thread performance detection tool, multi-threading may have thread-related performance problems such as load ratio balance, excessive synchronization overhead, etc. This tool can help you find the status of each thread at every moment.

2. Installation

On linux, decompress the file vtune_amplifier_xe_2011_update1.tar.gz;

Execute the install.sh script and follow the prompts to the next step;

Install a Windows client, VTune_Amplifier_XE_2011_update3_setup.exe, to display the collected data in a graphical interface.

3. Use

VTune has two working modes, we generally use the remote mode,

Local mode is when the program under analysis and VTune work on the same machine at the same time

Remote mode means that the program is on one machine and VTune is installed on another machine

1) Start the application that needs to be analyzed, taking is_searcher_server as an example

/home/admin/intel/vtune_amplifier_xe_2011/bin64/amplxe-cl -collect hotspots is_searcher_server -c conf/searcher_server.cfg -l conf/searcher_log.cfg -k restart -d

Among them, amplxe-cl refers to list the analysis types supported by this machine. Here, select -collect hotspots to collect hotspots.

There is also amplxe-cl --report-list; for supported report types.

更多见http://software.intel.com/zh-cn/blogs/2010/11/10/amplxe-cl/ 

后面就是服务的启动参数。

服务启动后,会自动生成一个数据文件,名为r000hs(之后生成的会自动命名为r001hs,r002hs…)

2)启动其它服务程序,在施压一段时间后,停止被分析的程序(也就是这里的is_searcher_server)

这时,r000hs里存的就是收集的数据信息。要注意,一定要停止被分析程序,否则数据不会生成到r000hs里面。

3)启动windows客户端VTune_Amplifier_XE_2011_update3_setup,将数据r000hs导入,步骤如下:

file->open->result,将r000hs加载,但此时会很慢,解决办法,使用sumba共享服务器,将linux文件在windows下共享

原文:http://ju.outofmemory.cn/entry/1506

Guess you like

Origin http://43.154.161.224:23101/article/api/json?id=325901454&siteId=291194637
Recommended