Visible to Intel only — GUID: GUID-0202D21D-52AE-45BE-9F41-AE38D412B053
Visible to Intel only — GUID: GUID-0202D21D-52AE-45BE-9F41-AE38D412B053
Linux* Targets
Use the Intel® VTune™ Profiler for performance analysis on local and remote Linux* target systems.
To analyze your Linux target, do the following:
Prepare your target application for analysis:
Enable downloading debug information for system kernels by installing debug info packages available for your system version.
Enable downloading debug information for the application binaries by using the -g option when compiling your code. Consider using the recommended compiler settings to make the performance analysis more effective.
Build your target in the Release mode.
Create a baseline against which you can compare the performance improvements as a result of tuning.
For example, you instrument your code to determine how long it takes to compress a certain file. Your original target code, augmented to provide these timing data, serves as your performance baseline. Every time you modify your target, compare the performance metrics of your optimized target with the baseline, to verify that the performance has improved.
Prepare your target system for analysis:
Build and install the sampling drivers, if required.
NOTE:- If the drivers were not built and set up during installation (for example, lack of privileges, missing kernel development RPM, and so on), VTune Profiler provides an error message and enables driverless sampling data collection based on the Linux Perf* tool functionality, which has a limited scope of analysis options.
- On Ubuntu* systems, VTune Profiler may fail to collect Hotspots and Threading analysis data if the scope of the ptrace() system call application is limited.
To workaround this issue for one session, set the value of the kernel.yama.ptrace_scopesysctl option to 0 with this command:
sysctl -w kernel.yama.ptrace_scope=0
To make this change permanent, see the corresponding Troubleshooting topic.
For remote analysis, configure SSH connection and set up your remote Linux system depending on the analysis usage mode.
Create a VTune Profiler project and run the performance analysis of your choice.