The btrecord and btreplay tools provide the ability to
record and replay IOs captured by the blktrace utility. Attempts
are made to maintain ordering, CPU mappings and time-separation of IOs.
The blktrace utility provides the ability to collect detailed
traces from the kernel for each IO processed by the block IO layer. The
traces provide a complete timeline for each IO processed, including
detailed information concerning when an IO was first received by the block
IO layer --- indicating the device, CPU number, time stamp, IO direction,
sector number and IO size (number of sectors). Using this information,
one is able to replay the IO again on the same machine or another
set up entirely.
The basic operating work-flow to replay IOs would be something like:
Run blktrace to collect traces. Here you specify the
device or devices that you wish to trace and later replay IOs upon. Note:
the only traces you are interested in are QUEUE requests ---
thus, to save system resources (including storage for traces), one could
specify the -a queue command line option to blktrace.
While blktrace is running, you run the workload that you
are interested in.
When the work load has completed, you stop the blktrace utility (thus saving all traces over the complete workload).
You extract the pertinent IO information from the traces saved by
blktrace using the btrecord utility. This will parse
each trace file created by blktrace, and crafty IO descriptions
to be used in the next phase of the workload processing.
Once btrecord has successfully created a series of data
files to be processed, you can run the btreplay utility which
attempts to generate the same IOs seen during the sample workload phase.
Set number of CPUs to use.
Set input directory.
This option requires a single parameter providing the directory
name for where input files are to be found. The default directory is the
current directory (.).
Find record files automatically
This option instructs btreplay to go find all the record files in the
directory specified (either via the -d option, or in the default
Show help and exit.
Set base name for input files.
Each input file has 3 fields:
Device identifier (taken directly from the device name of the
blktrace output file).
btrecord base name --- by default ``replay''.
The CPU number (again, taken directly from the
blktrace output file name).
This option requires a single parameter that will override the default name
(replay), and replace it with the specified value.
Set number of iterations to run.
This option requires a single parameter which specifies the number of times
to run through the input files. The default value is 1
Specify device mappings.
This option requires a single parameter which specifies the name of a
file contain device mappings. The file must be very simply managed, with
just two pieces of data per line:
The device name on the recorded system (with the '/dev/'
removed). Example: /dev/sda would just be sda.
The device name on the replay system to use (again, without the
'/dev/' path prepended).
An example file for when one would map devices /dev/sda and
/dev/sdb on the recorded system to dev/sdg and
sdh on the replay system would be:
The only entries in the file that are allowed are these two element lines ---
we do not (yet?) support the notion of blank lines, or comment lines, or the
The utility allows for multiple -M options to be
supplied on the command line.
Disable pre-bunch stalls.
When specified on the command line, all pre-bunch stall indicators will be
ignored. IOs will be replayed without inter-bunch delays.
Enable verbose output.
When specified on the command line, this option instructs btreplay
to store information concerning each stall and IO operation
performed by btreplay. The name of each file so created will be
the input file name used with an extension of .rep appended onto
it. Thus, an input file of the name sdab.replay.3 would generate a
verbose output file with the name sdab.replay.3.rep in the
directory specified for input files.
In addition, btreplay will also output to stderr the
names of the input files being processed.
Show version number and exit.
Enable writing during replay.
As a precautionary measure, by default btreplay will not
process write requests. In order to enable btreplay to
actually write to devices one must explicitly specify the
btreplay was written by Alan D. Brunelle. This
man page was created from the btreplay documentation by Bas Zoetekouw.