 | Level: Introductory Rob Nava (navar@us.ibm.com), Linux Performance Analyst, IBM
21 Jun 2006 Profilers help pinpoint common performance problems in an application. This article compares and contrasts three commonly used open source profilers for Linux™ on POWER™ -- OProfile, gprof, and Tprof -- which are available to end users and programmers for both SUSE™ and Red Hat® Linux distributions. To demonstrate each profiler's strengths and weaknesses, including any overhead the profilers add during runtime, this article profiles a simple sort program, incorporating three different sorting algorithms.
Introduction
Teams conducting performance analysis on Linux on POWER often have little practical experience with profilers other than their preferred profiling tool. Some use OProfile, which allows not only normal cycle event profiling, but also profiling based on any of the hardware performance counters. Other teams are very familiar with gprof and its capability to compile profiling hooks into the application. And still other teams prefer to use the Tprof profiler.
By comparing and contrasting these three system profilers, this article (see the Download section, below) shows which uses are possible and appropriate, the intent being to demonstrate each profiler's strengths and weaknesses.
The example workload is a simple sorting application, written in C, that has known performance issues when using different sorting algorithms. Each of the sorting algorithms is encapsulated in its own function, which makes it easy for a profiler to highlight the differences in the sorting functions.
Download | Description | Name | Size | Download method |
|---|
| Article in PDF format | l-profilers.pdf | 137KB | HTTP |
|---|
About the author | | | Rob is a Linux performance analyst based in Austin, Texas. Rob's main role is to improve Linux application and kernel performance on POWER and x86 architectures. Rob has also worked in the AIX Performance group. |
Rate this content
| | | |
