Linux system administrators regularly get complaints from the users that their system is (too) slow. When such complaint is serious, it might be difficult to determine why these delays happened, especially if the delay has been noticed in the past and seems to be an incident.
Many tools, like vmstat, iostat and top, only offer 'live' counters and should be running at the moment that the delays are experienced. Therefore, these tools are mainly useful in case of a structural lack of performance.
Other tools, like sar, maintain a permanent log to view the history. However, these tools only register system-level counters, so you can observe that a hardware resource is overloaded but have no idea about the process that overloads the resource.

In this presentation, I will discuss the steps that are needed during the performance analysis phase and I will highlight the hardware components that might cause the performance complaints. Furthermore, I will show how the load on these hardware components can be measured with the open source program 'atop', and in what respect 'atop' differs from similar performance analysis tools.