====== Block Layer I/O Tracing ======

===== Tools =====

Install //blktrace// on Debian based systems:

<cli prompt='# '>
root@host:~# apt-get install blktrace
</cli>

Ensure that the virtual filesystem Debug-FS is mounted:

<cli prompt='# '>
root@host:~# mount | grep debugfs
</cli>

Mount the Debug-FS if it's currently not available:

<cli prompt='# '>
root@host:~# /etc/init.d/mountdebugfs start
</cli>

or:

<cli prompt='# '>
root@host:~# mount -t debugfs debugfs /sys/kernel/debug
</cli>

Start the recording of the block I/O on a device until <kbd>CTRL+C</kbd> is pressed. In this example the block device ''/dev/sda'':

<cli prompt='# '>
root@host:~# blktrace /dev/sda
^C
=== sda ===
  CPU  0:                 8674 events,      407 KiB data
  CPU  1:                 9747 events,      457 KiB data
  CPU  2:                  631 events,       30 KiB data
  CPU  3:                 8977 events,      421 KiB data
  Total:                 28029 events (dropped 0),     1314 KiB data
</cli>

This creates the output files ''sda.blktrace.<CPU-#>'' in the current working directory. The output files can be analysed with the command:

<cli prompt='# '>
root@host:~# blkparse -i sda.blktrace.<CPU-#>
</cli>

See ''man blkparse'' for a description of the output format.

Recording and analysis of the block I/O on a device can be pipelined:

<cli prompt='# '>
root@host:~# blktrace /dev/sda -o - | blkparse -i -
</cli>

or the shorthand command:

<cli prompt='# '>
root@host:~# btrace /dev/sda
</cli>

be used.

To record and replay the block I/O on a device the commands ''btrecord'' and ''btreplay'' can be used. This can be especially useful to compare different environments or settings with the exact same block I/O load.

The tool ''btt'' perfoms an analysis on the output files written by the ''blktrace'' command:

<cli prompt='# '>
root@host:~# btt -i sda.blktrace.0
</cli>

See ''man btt'' for the available options.

The tool ''blkiomon'' can be used to perfom an iostat''-like analysis ''on the output generated by the ''blktrace'' command. For example to generate an 1 hour long trace of the device ''/dev/sda'' and output the statistics in a human-readable format in 10 second intervals: 

<cli prompt='# '>
root@host:~# blktrace /dev/sda -w 3600 -o - | blkiomon -I 10 -h -
</cli>

See ''man blkiomon'' for the available options.

For visualization purposes there are the tools ''bno_plot'' and ''iowatcher'' available.

===== Links =====

[[https://www.linux-magazin.de/ausgaben/2019/03/best-practices-8/|Linux Magazine - Issue 03/2019 - Plattenzugriffe bis aufs letzte Bit nachvollziehen]]\\