I/O tracing using blktrace

Edit online

The Linux kernel can collect events about all state changes of I/O requests. Later, the blktrace utilities can derive data from these events.

Before you begin

I/O tracing with blktrace requires two parts:
  • A Linux® kernel with the config option CONFIG_BLK_DEV_IO_TRACE enabled.
  • The blktrace userspace utilities, available from: 
    git://git.kernel.dk/blktrace.git
    or 
    http://brick.kernel.dk/snaps/
    The blktrace README file tells you where to get the sources, how to use blktrace and where to find the documentation.

About this task

You can collect data about I/O requests with the help of blktrace (see Figure 1).

Hint: While the I/O analysis can be run at the system where I/O is actually being traced, two Linux systems should be used: One that is being traced and the trace data is being redirected through a network connection to the second one for evaluation. This minimizes the impact on the system being traced.
Figure 1. Latency reported by blktrace

The graphic describes the parts of blktrace latency