HOWTO

Table of contents
-----------------

1. Overview
2. How fio works
3. Running fio
4. Job file format
5. Detailed list of parameters
6. Normal output
7. Terse output
8. Trace file format
9. CPU idleness profiling
10. Verification and triggers

1.0 Overview and history
------------------------
fio was originally written to save me the hassle of writing special test
case programs when I wanted to test a specific workload, either for
performance reasons or to find/reproduce a bug. The process of writing
such a test app can be tiresome, especially if you have to do it often.
Hence I needed a tool that would be able to simulate a given io workload
without resorting to writing a tailored test case again and again.

A test work load is difficult to define, though. There can be any number
of processes or threads involved, and they can each be using their own
way of generating io. You could have someone dirtying large amounts of
memory in an memory mapped file, or maybe several threads issuing
reads using asynchronous io. fio needed to be flexible enough to
simulate both of these cases, and many more.

2.0 How fio works
-----------------
The first step in getting fio to simulate a desired io workload, is
writing a job file describing that specific setup. A job file may contain
any number of threads and/or files - the typical contents of the job file
is a global section defining shared parameters, and one or more job
sections describing the jobs involved. When run, fio parses this file
and sets everything up as described. If we break down a job from top to
bottom, it contains the following basic parameters:

	IO type		Defines the io pattern issued to the file(s).
			We may only be reading sequentially from this
			file(s), or we may be writing randomly. Or even
			mixing reads and writes, sequentially or randomly.

	Block size	In how large chunks are we issuing io? This may be
			a single value, or it may describe a range of
			block sizes.

	IO size		How much data are we going to be reading/writing.

	IO engine	How do we issue io? We could be memory mapping the
			file, we could be using regular read/write, we
			could be using splice, async io, syslet, or even
			SG (SCSI generic sg).

	IO depth	If the io engine is async, how large a queuing
			depth do we want to maintain?

	IO type		Should we be doing buffered io, or direct/raw io?

	Num files	How many files are we spreading the workload over.

	Num threads	How many threads or processes should we spread
			this workload over.

The above are the basic parameters defined for a workload, in addition
there's a multitude of parameters that modify other aspects of how this
job behaves.


3.0 Running fio
---------------
See the README file for command line parameters, there are only a few
of them.

Running fio is normally the easiest part - you just give it the job file
(or job files) as parameters:

$ fio job_file

and it will start doing what the job_file tells it to do. You can give
more than one job file on the command line, fio will serialize the running
of those files. Internally that is the same as using the 'stonewall'
parameter described in the parameter section.

If the job file contains only one job, you may as well just give the
parameters on the command line. The command line parameters are identical
to the job parameters, with a few extra that control global parameters
(see README). For example, for the job file parameter iodepth=2, the
mirror command line option would be --iodepth 2 or --iodepth=2. You can
also use the command line for giving more than one job entry. For each
--name option that fio sees, it will start a new job with that name.
Command line entries following a --name entry will apply to that job,
until there are no more entries or a new --name entry is seen. This is
similar to the job file options, where each option applies to the current
job until a new [] job entry is seen.

fio does not need to run as root, except if the files or devices specified
in the job section requires that. Some other options may also be restricted,
such as memory locking, io scheduler switching, and decreasing the nice value.


4.0 Job file format
-------------------
As previously described, fio accepts one or more job files describing
what it is supposed to do. The job file format is the classic ini file,
where the names enclosed in [] brackets define the job name. You are free
to use any ascii name you want, except 'global' which has special meaning.
A global section sets defaults for the jobs described in that file. A job
may override a global section parameter, and a job file may even have
several global sections if so desired. A job is only affected by a global
section residing above it. If the first character in a line is a ';' or a
'#', the entire line is discarded as a comment.

So let's look at a really simple job file that defines two processes, each
randomly reading from a 128MB file.

; -- start job file --
[global]
rw=randread
size=128m

[job1]

[job2]

; -- end job file --

As you can see, the job file sections themselves are empty as all the
described parameters are shared. As no filename= option is given, fio
makes up a filename for each of the jobs as it sees fit. On the command
line, this job would look as follows:

$ fio --name=global --rw=randread --size=128m --name=job1 --name=job2


Let's look at an example that has a number of processes writing randomly
to files.

; -- start job file --
[random-writers]
ioengine=libaio
iodepth=4
rw=randwrite
bs=32k
direct=0
size=64m
numjobs=4

; -- end job file --

Here we have no global section, as we only have one job defined anyway.
We want to use async io here, with a depth of 4 for each file. We also
increased the buffer size used to 32KB and define numjobs to 4 to
fork 4 identical jobs. The result is 4 processes each randomly writing
to their own 64MB file. Instead of using the above job file, you could
have given the parameters on the command line. For this case, you would
specify:

$ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4

When fio is utilized as a basis of any reasonably large test suite, it might be
desirable to share a set of standardized settings across multiple job files.
Instead of copy/pasting such settings, any section may pull in an external
.fio file with 'include filename' directive, as in the following example:

; -- start job file including.fio --
[global]
filename=/tmp/test
filesize=1m
include glob-include.fio

[test]
rw=randread
bs=4k
time_based=1
runtime=10
include test-include.fio
; -- end job file including.fio --

; -- start job file glob-include.fio --
thread=1
group_reporting=1
; -- end job file glob-include.fio --

; -- start job file test-include.fio --
ioengine=libaio
iodepth=4
; -- end job file test-include.fio --

Settings pulled into a section apply to that section only (except global
section). Include directives may be nested in that any included file may
contain further include directive(s). Include files may not contain []
sections.


4.1 Environment variables
-------------------------

fio also supports environment variable expansion in job files. Any
sub-string of the form "${VARNAME}" as part of an option value (in other
words, on the right of the `='), will be expanded to the value of the
environment variable called VARNAME.  If no such environment variable
is defined, or VARNAME is the empty string, the empty string will be
substituted.

As an example, let's look at a sample fio invocation and job file:

$ SIZE=64m NUMJOBS=4 fio jobfile.fio

; -- start job file --
[random-writers]
rw=randwrite
size=${SIZE}
numjobs=${NUMJOBS}
; -- end job file --

This will expand to the following equivalent job file at runtime:

; -- start job file --
[random-writers]
rw=randwrite
size=64m
numjobs=4
; -- end job file --

fio ships with a few example job files, you can also look there for
inspiration.

4.2 Reserved keywords
---------------------

Additionally, fio has a set of reserved keywords that will be replaced
internally with the appropriate value. Those keywords are:

$pagesize	The architecture page size of the running system
$mb_memory	Megabytes of total memory in the system
$ncpus		Number of online available CPUs

These can be used on the command line or in the job file, and will be
automatically substituted with the current system values when the job
is run. Simple math is also supported on these keywords, so you can
perform actions like:

size=8*$mb_memory

and get that properly expanded to 8 times the size of memory in the
machine.


5.0 Detailed list of parameters
-------------------------------

This section describes in details each parameter associated with a job.
Some parameters take an option of a given type, such as an integer or
a string. Anywhere a numeric value is required, an arithmetic expression
may be used, provided it is surrounded by parentheses. Supported operators
are:

	addition (+)
	subtraction (-)
	multiplication (*)
	division (/)
	modulus (%)
	exponentiation (^)

For time values in expressions, units are microseconds by default. This is
different than for time values not in expressions (not enclosed in
parentheses). The following types are used:

str	String. This is a sequence of alpha characters.
time	Integer with possible time suffix. In seconds unless otherwise
	specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds,
	minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds,
	and 'us' (or 'usec') for microseconds.
int	SI integer. A whole number value, which may contain a suffix
	describing the base of the number. Accepted suffixes are k/m/g/t/p,
	meaning kilo, mega, giga, tera, and peta. The suffix is not case
	sensitive, and you may also include trailing 'b' (eg 'kb' is the same
	as 'k'). So if you want to specify 4096, you could either write
	out '4096' or just give 4k. The suffixes signify base 2 values, so
	1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly
	set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the
	case, then 1000 is used as the multiplier. This can be handy for
	disks, since manufacturers generally use base 10 values when listing
	the capacity of a drive. If the option accepts an upper and lower
	range, use a colon ':' or minus '-' to separate such values.  May also
	include a prefix to indicate numbers base. If 0x is used, the number
	is assumed to be hexadecimal.  See irange.
bool	Boolean. Usually parsed as an integer, however only defined for
	true and false (1 and 0).
irange	Integer range with suffix. Allows value range to be given, such
	as 1024-4096. A colon may also be used as the separator, eg
	1k:4k. If the option allows two sets of ranges, they can be
	specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see
	int.
float_list	A list of floating numbers, separated by a ':' character.

With the above in mind, here follows the complete list of fio job
parameters.

name=str	ASCII name of the job. This may be used to override the
		name printed by fio for this job. Otherwise the job
		name is used. On the command line this parameter has the
		special purpose of also signaling the start of a new
		job.

wait_for=str	Specifies the name of the already defined job to wait
		for. Single waitee name only may be specified. If set, the job
		won't be started until all workers of the waitee job are done.

		Wait_for operates on the job name basis, so there are a few
		limitations. First, the waitee must be defined prior to the
		waiter job (meaning no forward references). Second, if a job
		is being referenced as a waitee, it must have a unique name
		(no duplicate waitees).

description=str	Text description of the job. Doesn't do anything except
		dump this text description when this job is run. It's
		not parsed.

directory=str	Prefix filenames with this directory. Used to place files
		in a different location than "./". See the 'filename' option
		for escaping certain characters.

filename=str	Fio normally makes up a filename based on the job name,
		thread number, and file number. If you want to share
		files between threads in a job or several jobs, specify
		a filename for each of them to override the default. If
		the ioengine used is 'net', the filename is the host, port,
		and protocol to use in the format of =host,port,protocol.
		See ioengine=net for more. If the ioengine is file based, you
		can specify a number of files by separating the names with a
		':' colon. So if you wanted a job to open /dev/sda and /dev/sdb
		as the two working files, you would use
		filename=/dev/sda:/dev/sdb. On Windows, disk devices are
		accessed as \\.\PhysicalDrive0 for the first device,
		\\.\PhysicalDrive1 for the second etc. Note: Windows and
		FreeBSD prevent write access to areas of the disk containing
		in-use data (e.g. filesystems).
		If the wanted filename does need to include a colon, then
		escape that with a '\' character. For instance, if the filename
		is "/dev/dsk/foo@3,0:c", then you would use
		filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning
		stdin or stdout. Which of the two depends on the read/write
		direction set.

filename_format=str
		If sharing multiple files between jobs, it is usually necessary
		to  have fio generate the exact names that you want. By default,
		fio will name a file based on the default file format
		specification of jobname.jobnumber.filenumber. With this
		option, that can be customized. Fio will recognize and replace
		the following keywords in this string:

		$jobname
			The name of the worker thread or process.

		$jobnum
			The incremental number of the worker thread or
			process.

		$filenum
			The incremental number of the file for that worker
			thread or process.

		To have dependent jobs share a set of files, this option can
		be set to have fio generate filenames that are shared between
		the two. For instance, if testfiles.$filenum is specified,
		file number 4 for any job will be named testfiles.4. The
		default of $jobname.$jobnum.$filenum will be used if
		no other format specifier is given.

opendir=str	Tell fio to recursively add any file it can find in this
		directory and down the file system tree.

lockfile=str	Fio defaults to not locking any files before it does
		IO to them. If a file or file descriptor is shared, fio
		can serialize IO to that file to make the end result
		consistent. This is usual for emulating real workloads that
		share files. The lock modes are:

			none		No locking. The default.
			exclusive	Only one thread/process may do IO,
					excluding all others.
			readwrite	Read-write locking on the file. Many
					readers may access the file at the
					same time, but writes get exclusive
					access.

readwrite=str
rw=str		Type of io pattern. Accepted values are:

			read		Sequential reads
			write		Sequential writes
			randwrite	Random writes
			randread	Random reads
			rw,readwrite	Sequential mixed reads and writes
			randrw		Random mixed reads and writes
			trimwrite	Mixed trims and writes. Blocks will be
					trimmed first, then written to.

		For the mixed io types, the default is to split them 50/50.
		For certain types of io the result may still be skewed a bit,
		since the speed may be different. It is possible to specify
		a number of IO's to do before getting a new offset, this is
		done by appending a ':<nr>' to the end of the string given.
		For a random read, it would look like 'rw=randread:8' for
		passing in an offset modifier with a value of 8. If the
		suffix is used with a sequential IO pattern, then the value
		specified will be added to the generated offset for each IO.
		For instance, using rw=write:4k will skip 4k for every
		write. It turns sequential IO into sequential IO with holes.
		See the 'rw_sequencer' option.

rw_sequencer=str If an offset modifier is given by appending a number to
		the rw=<str> line, then this option controls how that
		number modifies the IO offset being generated. Accepted
		values are:

			sequential	Generate sequential offset
			identical	Generate the same offset

		'sequential' is only useful for random IO, where fio would
		normally generate a new random offset for every IO. If you
		append eg 8 to randread, you would get a new random offset for
		every 8 IO's. The result would be a seek for only every 8
		IO's, instead of for every IO. Use rw=randread:8 to specify
		that. As sequential IO is already sequential, setting
		'sequential' for that would not result in any differences.
		'identical' behaves in a similar fashion, except it sends
		the same offset 8 number of times before generating a new
		offset.

kb_base=int	The base unit for a kilobyte. The defacto base is 2^10, 1024.
		Storage manufacturers like to use 10^3 or 1000 as a base
		ten unit instead, for obvious reasons. Allow values are
		1024 or 1000, with 1024 being the default.

unified_rw_reporting=bool	Fio normally reports statistics on a per
		data direction basis, meaning that read, write, and trim are
		accounted and reported separately. If this option is set,
		the fio will sum the results and report them as "mixed"
		instead.

randrepeat=bool	For random IO workloads, seed the generator in a predictable
		way so that results are repeatable across repetitions.
		Defaults to true.

randseed=int	Seed the random number generators based on this seed value, to
		be able to control what sequence of output is being generated.
		If not set, the random sequence depends on the randrepeat
		setting.

fallocate=str	Whether pre-allocation is performed when laying down files.
		Accepted values are:

			none		Do not pre-allocate space
			posix		Pre-allocate via posix_fallocate()
			keep		Pre-allocate via fallocate() with
					FALLOC_FL_KEEP_SIZE set
			0		Backward-compatible alias for 'none'
			1		Backward-compatible alias for 'posix'

		May not be available on all supported platforms. 'keep' is only
		available on Linux.If using ZFS on Solaris this must be set to
		'none' because ZFS doesn't support it. Default: 'posix'.

fadvise_hint=bool By default, fio will use fadvise() to advise the kernel
		on what IO patterns it is likely to issue. Sometimes you
		want to test specific IO patterns without telling the
		kernel about it, in which case you can disable this option.
		If set, fio will use POSIX_FADV_SEQUENTIAL for sequential
		IO and POSIX_FADV_RANDOM for random IO.

fadvise_stream=int Notify the kernel what write stream ID to place these
		writes under. Only supported on Linux. Note, this option
		may change going forward.

size=int	The total size of file io for this job. Fio will run until
		this many bytes has been transferred, unless runtime is
		limited by other options (such as 'runtime', for instance,
		or increased/decreased by 'io_size'). Unless specific nrfiles
		and filesize options are given, fio will divide this size
		between the available files specified by the job. If not set,
		fio will use the full size of the given files or devices.
		If the files do not exist, size must be given. It is also
		possible to give size as a percentage between 1 and 100. If
		size=20% is given, fio will use 20% of the full size of the
		given files or devices.

io_size=int
io_limit=int	Normally fio operates within the region set by 'size', which
		means that the 'size' option sets both the region and size of
		IO to be performed. Sometimes that is not what you want. With
		this option, it is possible to define just the amount of IO
		that fio should do. For instance, if 'size' is set to 20G and
		'io_size' is set to 5G, fio will perform IO within the first
		20G but exit when 5G have been done. The opposite is also
		possible - if 'size' is set to 20G, and 'io_size' is set to
		40G, then fio will do 40G of IO within the 0..20G region.

filesize=int	Individual file sizes. May be a range, in which case fio
		will select sizes for files at random within the given range
		and limited to 'size' in total (if that is given). If not
		given, each created file is the same size.

file_append=bool	Perform IO after the end of the file. Normally fio will
		operate within the size of a file. If this option is set, then
		fio will append to the file instead. This has identical
		behavior to setting offset to the size of a file. This option
		is ignored on non-regular files.

fill_device=bool
fill_fs=bool	Sets size to something really large and waits for ENOSPC (no
		space left on device) as the terminating condition. Only makes
		sense with sequential write. For a read workload, the mount
		point will be filled first then IO started on the result. This
		option doesn't make sense if operating on a raw device node,
		since the size of that is already known by the file system.
		Additionally, writing beyond end-of-device will not return
		ENOSPC there.

blocksize=int
bs=int		The block size used for the io units. Defaults to 4k. Values
		can be given for both read and writes. If a single int is
		given, it will apply to both. If a second int is specified
		after a comma, it will apply to writes only. In other words,
		the format is either bs=read_and_write or bs=read,write,trim.
		bs=4k,8k will thus use 4k blocks for reads, 8k blocks for
		writes, and 8k for trims. You can terminate the list with
		a trailing comma. bs=4k,8k, would use the default value for
		trims.. If you only wish to set the write size, you
		can do so by passing an empty read size - bs=,8k will set
		8k for writes and leave the read default value.

blockalign=int
ba=int		At what boundary to align random IO offsets. Defaults to
		the same as 'blocksize' the minimum blocksize given.
		Minimum alignment is typically 512b for using direct IO,
		though it usually depends on the hardware block size. This
		option is mutually exclusive with using a random map for
		files, so it will turn off that option.

blocksize_range=irange
bsrange=irange	Instead of giving a single block size, specify a range
		and fio will mix the issued io block sizes. The issued
		io unit will always be a multiple of the minimum value
		given (also see bs_unaligned). Applies to both reads and
		writes, however a second range can be given after a comma.
		See bs=.

bssplit=str	Sometimes you want even finer grained control of the
		block sizes issued, not just an even split between them.
		This option allows you to weight various block sizes,
		so that you are able to define a specific amount of
		block sizes issued. The format for this option is:

			bssplit=blocksize/percentage:blocksize/percentage

		for as many block sizes as needed. So if you want to define
		a workload that has 50% 64k blocks, 10% 4k blocks, and
		40% 32k blocks, you would write:

			bssplit=4k/10:64k/50:32k/40

		Ordering does not matter. If the percentage is left blank,
		fio will fill in the remaining values evenly. So a bssplit
		option like this one:

			bssplit=4k/50:1k/:32k/

		would have 50% 4k ios, and 25% 1k and 32k ios. The percentages
		always add up to 100, if bssplit is given a range that adds
		up to more, it will error out.

		bssplit also supports giving separate splits to reads and
		writes. The format is identical to what bs= accepts. You
		have to separate the read and write parts with a comma. So
		if you want a workload that has 50% 2k reads and 50% 4k reads,
		while having 90% 4k writes and 10% 8k writes, you would
		specify:

		bssplit=2k/50:4k/50,4k/90:8k/10

blocksize_unaligned
bs_unaligned	If this option is given, any byte size value within bsrange
		may be used as a block range. This typically wont work with
		direct IO, as that normally requires sector alignment.

bs_is_seq_rand	If this option is set, fio will use the normal read,write
		blocksize settings as sequential,random instead. Any random
		read or write will use the WRITE blocksize settings, and any
		sequential read or write will use the READ blocksize setting.

zero_buffers	If this option is given, fio will init the IO buffers to
		all zeroes. The default is to fill them with random data.

refill_buffers	If this option is given, fio will refill the IO buffers
		on every submit. The default is to only fill it at init
		time and reuse that data. Only makes sense if zero_buffers
		isn't specified, naturally. If data verification is enabled,
		refill_buffers is also automatically enabled.

scramble_buffers=bool	If refill_buffers is too costly and the target is
		using data deduplication, then setting this option will
		slightly modify the IO buffer contents to defeat normal
		de-dupe attempts. This is not enough to defeat more clever
		block compression attempts, but it will stop naive dedupe of
		blocks. Default: true.

buffer_compress_percentage=int	If this is set, then fio will attempt to
		provide IO buffer content (on WRITEs) that compress to
		the specified level. Fio does this by providing a mix of
		random data and a fixed pattern. The fixed pattern is either
		zeroes, or the pattern specified by buffer_pattern. If the
		pattern option is used, it might skew the compression ratio
		slightly. Note that this is per block size unit, for file/disk
		wide compression level that matches this setting, you'll also
		want to set refill_buffers.

buffer_compress_chunk=int	See buffer_compress_percentage. This
		setting allows fio to manage how big the ranges of random
		data and zeroed data is. Without this set, fio will
		provide buffer_compress_percentage of blocksize random
		data, followed by the remaining zeroed. With this set
		to some chunk size smaller than the block size, fio can
		alternate random and zeroed data throughout the IO
		buffer.

buffer_pattern=str	If set, fio will fill the io buffers with this
		pattern. If not set, the contents of io buffers is defined by
		the other options related to buffer contents. The setting can
		be any pattern of bytes, and can be prefixed with 0x for hex
		values. It may also be a string, where the string must then
		be wrapped with "", e.g.:

		buffer_pattern="abcd"
		  or
		buffer_pattern=-12
		  or
		buffer_pattern=0xdeadface

		Also you can combine everything together in any order:
		buffer_pattern=0xdeadface"abcd"-12

dedupe_percentage=int	If set, fio will generate this percentage of
		identical buffers when writing. These buffers will be
		naturally dedupable. The contents of the buffers depend on
		what other buffer compression settings have been set. It's
		possible to have the individual buffers either fully
		compressible, or not at all. This option only controls the
		distribution of unique buffers.

nrfiles=int	Number of files to use for this job. Defaults to 1.

openfiles=int	Number of files to keep open at the same time. Defaults to
		the same as nrfiles, can be set smaller to limit the number
		simultaneous opens.

file_service_type=str  Defines how fio decides which file from a job to
		service next. The following types are defined:

			random	Just choose a file at random.

			roundrobin  Round robin over open files. This
				is the default.

			sequential  Finish one file before moving on to
				the next. Multiple files can still be
				open depending on 'openfiles'.

		The string can have a number appended, indicating how
		often to switch to a new file. So if option random:4 is
		given, fio will switch to a new random file after 4 ios
		have been issued.

ioengine=str	Defines how the job issues io to the file. The following
		types are defined:

			sync	Basic read(2) or write(2) io. lseek(2) is
				used to position the io location.

			psync 	Basic pread(2) or pwrite(2) io.

			vsync	Basic readv(2) or writev(2) IO.

			psyncv	Basic preadv(2) or pwritev(2) IO.

			libaio	Linux native asynchronous io. Note that Linux
				may only support queued behaviour with
				non-buffered IO (set direct=1 or buffered=0).
				This engine defines engine specific options.

			posixaio glibc posix asynchronous io.

			solarisaio Solaris native asynchronous io.

			windowsaio Windows native asynchronous io.

			mmap	File is memory mapped and data copied
				to/from using memcpy(3).

			splice	splice(2) is used to transfer the data and
				vmsplice(2) to transfer data from user
				space to the kernel.

			syslet-rw Use the syslet system calls to make
				regular read/write async.

			sg	SCSI generic sg v3 io. May either be
				synchronous using the SG_IO ioctl, or if
				the target is an sg character device
				we use read(2) and write(2) for asynchronous
				io.

			null	Doesn't transfer any data, just pretends
				to. This is mainly used to exercise fio
				itself and for debugging/testing purposes.

			net	Transfer over the network to given host:port.
				Depending on the protocol used, the hostname,
				port, listen and filename options are used to
				specify what sort of connection to make, while
				the protocol option determines which protocol
				will be used.
				This engine defines engine specific options.

			netsplice Like net, but uses splice/vmsplice to
				map data and send/receive.
				This engine defines engine specific options.

			cpuio	Doesn't transfer any data, but burns CPU
				cycles according to the cpuload= and
				cpucycle= options. Setting cpuload=85
				will cause that job to do nothing but burn
				85% of the CPU. In case of SMP machines,
				use numjobs=<no_of_cpu> to get desired CPU
				usage, as the cpuload only loads a single
				CPU at the desired rate.

			guasi	The GUASI IO engine is the Generic Userspace
				Asyncronous Syscall Interface approach
				to async IO. See

				http://www.xmailserver.org/guasi-lib.html

				for more info on GUASI.

			rdma    The RDMA I/O engine  supports  both  RDMA
				memory semantics (RDMA_WRITE/RDMA_READ) and
				channel semantics (Send/Recv) for the
				InfiniBand, RoCE and iWARP protocols.

			falloc	IO engine that does regular fallocate to
				simulate data transfer as fio ioengine.
				DDIR_READ  does fallocate(,mode = keep_size,)
				DDIR_WRITE does fallocate(,mode = 0)
				DDIR_TRIM  does fallocate(,mode = punch_hole)

			e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT
				ioctls to simulate defragment activity in
				request to DDIR_WRITE event

			rbd	IO engine supporting direct access to Ceph
				Rados Block Devices (RBD) via librbd without
				the need to use the kernel rbd driver. This
				ioengine defines engine specific options.

			gfapi	Using Glusterfs libgfapi sync interface to
				direct access to Glusterfs volumes without
				options.

			gfapi_async Using Glusterfs libgfapi async interface
				to direct access to Glusterfs volumes without
				having to go through FUSE. This ioengine
				defines engine specific options.

			libhdfs	Read and write through Hadoop (HDFS).
				This engine interprets offsets a little
				differently. In HDFS, files once created
				cannot be modified. So random writes are not
				possible. To imitate this, libhdfs engine
				creates bunch of small files, and engine will
				pick a file out of those files based on the 
				offset enerated by fio backend. Each jobs uses
				it's own connection to HDFS.

			mtd	Read, write and erase an MTD character device
				(e.g., /dev/mtd0). Discards are treated as
				erases. Depending on the underlying device
				type, the I/O may have to go in a certain
				pattern, e.g., on NAND, writing sequentially
				to erase blocks and discarding before
				overwriting. The writetrim mode works well
				for this constraint.

			external Prefix to specify loading an external
				IO engine object file. Append the engine
				filename, eg ioengine=external:/tmp/foo.o
				to load ioengine foo.o in /tmp.

iodepth=int	This defines how many io units to keep in flight against
		the file. The default is 1 for each file defined in this
		job, can be overridden with a larger value for higher
		concurrency. Note that increasing iodepth beyond 1 will not
		affect synchronous ioengines (except for small degress when
		verify_async is in use). Even async engines may impose OS
		restrictions causing the desired depth not to be achieved.
		This may happen on Linux when using libaio and not setting
		direct=1, since buffered IO is not async on that OS. Keep an
		eye on the IO depth distribution in the fio output to verify
		that the achieved depth is as expected. Default: 1.

iodepth_batch_submit=int
iodepth_batch=int This defines how many pieces of IO to submit at once.
		It defaults to 1 which means that we submit each IO
		as soon as it is available, but can be raised to submit
		bigger batches of IO at the time. If it is set to 0 the iodepth
		value will be used.

iodepth_batch_complete_min=int
iodepth_batch_complete=int This defines how many pieces of IO to retrieve
		at once. It defaults to 1 which means that we'll ask
		for a minimum of 1 IO in the retrieval process from
		the kernel. The IO retrieval will go on until we
		hit the limit set by iodepth_low. If this variable is
		set to 0, then fio will always check for completed
		events before queuing more IO. This helps reduce
		IO latency, at the cost of more retrieval system calls.

iodepth_batch_complete_max=int This defines maximum pieces of IO to
		retrieve at once. This variable should be used along with
		iodepth_batch_complete_min=int variable, specifying the range
		of min and max amount of IO which should be retrieved. By default
		it is equal to iodepth_batch_complete_min value.

		Example #1:

		iodepth_batch_complete_min=1
		iodepth_batch_complete_max=<iodepth>

		which means that we will retrieve at leat 1 IO and up to the
		whole submitted queue depth. If none of IO has been completed
		yet, we will wait.

		Example #2:

		iodepth_batch_complete_min=0
		iodepth_batch_complete_max=<iodepth>

		which means that we can retrieve up to the whole submitted
		queue depth, but if none of IO has been completed yet, we will
		NOT wait and immediately exit the system call. In this example
		we simply do polling.

iodepth_low=int	The low water mark indicating when to start filling
		the queue again. Defaults to the same as iodepth, meaning
		that fio will attempt to keep the queue full at all times.
		If iodepth is set to eg 16 and iodepth_low is set to 4, then
		after fio has filled the queue of 16 requests, it will let
		the depth drain down to 4 before starting to fill it again.

io_submit_mode=str	This option controls how fio submits the IO to
		the IO engine. The default is 'inline', which means that the
		fio job threads submit and reap IO directly. If set to
		'offload', the job threads will offload IO submission to a
		dedicated pool of IO threads. This requires some coordination
		and thus has a bit of extra overhead, especially for lower
		queue depth IO where it can increase latencies. The benefit
		is that fio can manage submission rates independently of
		the device completion rates. This avoids skewed latency
		reporting if IO gets back up on the device side (the
		coordinated omission problem).

direct=bool	If value is true, use non-buffered io. This is usually
		O_DIRECT. Note that ZFS on Solaris doesn't support direct io.
		On Windows the synchronous ioengines don't support direct io.

atomic=bool	If value is true, attempt to use atomic direct IO. Atomic
		writes are guaranteed to be stable once acknowledged by
		the operating system. Only Linux supports O_ATOMIC right
		now.

buffered=bool	If value is true, use buffered io. This is the opposite
		of the 'direct' option. Defaults to true.

offset=int	Start io at the given offset in the file. The data before
		the given offset will not be touched. This effectively
		caps the file size at real_size - offset.

offset_increment=int	If this is provided, then the real offset becomes
		offset + offset_increment * thread_number, where the thread
		number is a counter that starts at 0 and is incremented for
		each sub-job (i.e. when numjobs option is specified). This
		option is useful if there are several jobs which are intended
		to operate on a file in parallel disjoint segments, with
		even spacing between the starting points.

number_ios=int	Fio will normally perform IOs until it has exhausted the size
		of the region set by size=, or if it exhaust the allocated
		time (or hits an error condition). With this setting, the
		range/size can be set independently of the number of IOs to
		perform. When fio reaches this number, it will exit normally
		and report status. Note that this does not extend the amount
		of IO that will be done, it will only stop fio if this
		condition is met before other end-of-job criteria.

fsync=int	If writing to a file, issue a sync of the dirty data
		for every number of blocks given. For example, if you give
		32 as a parameter, fio will sync the file for every 32
		writes issued. If fio is using non-buffered io, we may
		not sync the file. The exception is the sg io engine, which
		synchronizes the disk cache anyway.

fdatasync=int	Like fsync= but uses fdatasync() to only sync data and not
		metadata blocks.
		In FreeBSD and Windows there is no fdatasync(), this falls back
		to using fsync()

sync_file_range=str:val	Use sync_file_range() for every 'val' number of
		write operations. Fio will track range of writes that
		have happened since the last sync_file_range() call. 'str'
		can currently be one or more of:

		wait_before	SYNC_FILE_RANGE_WAIT_BEFORE
		write		SYNC_FILE_RANGE_WRITE
		wait_after	SYNC_FILE_RANGE_WAIT_AFTER

		So if you do sync_file_range=wait_before,write:8, fio would
		use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for
		every 8 writes. Also see the sync_file_range(2) man page.
		This option is Linux specific.

overwrite=bool	If true, writes to a file will always overwrite existing
		data. If the file doesn't already exist, it will be
		created before the write phase begins. If the file exists
		and is large enough for the specified write phase, nothing
		will be done.

end_fsync=bool	If true, fsync file contents when a write stage has completed.

fsync_on_close=bool	If true, fio will fsync() a dirty file on close.
		This differs from end_fsync in that it will happen on every
		file close, not just at the end of the job.

rwmixread=int	How large a percentage of the mix should be reads.

rwmixwrite=int	How large a percentage of the mix should be writes. If both
		rwmixread and rwmixwrite is given and the values do not add
		up to 100%, the latter of the two will be used to override
		the first. This may interfere with a given rate setting,
		if fio is asked to limit reads or writes to a certain rate.
		If that is the case, then the distribution may be skewed.

random_distribution=str:float	By default, fio will use a completely uniform
		random distribution when asked to perform random IO. Sometimes
		it is useful to skew the distribution in specific ways,
		ensuring that some parts of the data is more hot than others.
		fio includes the following distribution models:

		random		Uniform random distribution
		zipf		Zipf distribution
		pareto		Pareto distribution
		gauss		Normal (guassian) distribution
		zoned		Zoned random distribution

		When using a zipf or pareto distribution, an input value
		is also needed to define the access pattern. For zipf, this
		is the zipf theta. For pareto, it's the pareto power. Fio
		includes a test program, genzipf, that can be used visualize
		what the given input values will yield in terms of hit rates.
		If you wanted to use zipf with a theta of 1.2, you would use
		random_distribution=zipf:1.2 as the option. If a non-uniform
		model is used, fio will disable use of the random map. For
		the gauss distribution, a normal deviation is supplied as
		a value between 0 and 100.

		For a zoned distribution, fio supports specifying percentages
		of IO access that should fall within what range of the file or
		device. For example, given a criteria of:

			60% of accesses should be to the first 10%
			30% of accesses should be to the next 20%
			8% of accesses should be to to the next 30%
			2% of accesses should be to the next 40%

		we can define that through zoning of the random accesses. For
		the above example, the user would do:

			random_distribution=zoned:60/10:30/20:8/30:2/40

		similarly to how bssplit works for setting ranges and
		percentages of block sizes. Like bssplit, it's possible to
		specify separate zones for reads, writes, and trims. If just
		one set is given, it'll apply to all of them.

percentage_random=int	For a random workload, set how big a percentage should
		be random. This defaults to 100%, in which case the workload
		is fully random. It can be set from anywhere from 0 to 100.
		Setting it to 0 would make the workload fully sequential. Any
		setting in between will result in a random mix of sequential
		and random IO, at the given percentages. It is possible to
		set different values for reads, writes, and trim. To do so,
		simply use a comma separated list. See blocksize.
	
norandommap	Normally fio will cover every block of the file when doing
		random IO. If this option is given, fio will just get a
		new random offset without looking at past io history. This
		means that some blocks may not be read or written, and that
		some blocks may be read/written more than once. If this option
		is used with verify= and multiple blocksizes (via bsrange=),
		only intact blocks are verified, i.e., partially-overwritten
		blocks are ignored.

softrandommap=bool See norandommap. If fio runs with the random block map
		enabled and it fails to allocate the map, if this option is
		set it will continue without a random block map. As coverage
		will not be as complete as with random maps, this option is
		disabled by default.

random_generator=str	Fio supports the following engines for generating
		IO offsets for random IO:

		tausworthe	Strong 2^88 cycle random number generator
		lfsr		Linear feedback shift register generator
		tausworthe64	Strong 64-bit 2^258 cycle random number
				generator

		Tausworthe is a strong random number generator, but it
		requires tracking on the side if we want to ensure that
		blocks are only read or written once. LFSR guarantees
		that we never generate the same offset twice, and it's
		also less computationally expensive. It's not a true
		random generator, however, though for IO purposes it's
		typically good enough. LFSR only works with single
		block sizes, not with workloads that use multiple block
		sizes. If used with such a workload, fio may read or write
		some blocks multiple times. The default value is tausworthe,
		unless the required space exceeds 2^32 blocks. If it does,
		then tausworthe64 is selected automatically.

nice=int	Run the job with the given nice value. See man nice(2).

prio=int	Set the io priority value of this job. Linux limits us to
		a positive value between 0 and 7, with 0 being the highest.
		See man ionice(1).

prioclass=int	Set the io priority class. See man ionice(1).

thinktime=int	Stall the job x microseconds after an io has completed before
		issuing the next. May be used to simulate processing being
		done by an application. See thinktime_blocks and
		thinktime_spin.

thinktime_spin=int
		Only valid if thinktime is set - pretend to spend CPU time
		doing something with the data received, before falling back
		to sleeping for the rest of the period specified by
		thinktime.

thinktime_blocks=int
		Only valid if thinktime is set - control how many blocks
		to issue, before waiting 'thinktime' usecs. If not set,
		defaults to 1 which will make fio wait 'thinktime' usecs
		after every block. This effectively makes any queue depth
		setting redundant, since no more than 1 IO will be queued
		before we have to complete it and do our thinktime. In
		other words, this setting effectively caps the queue depth
		if the latter is larger.

rate=int	Cap the bandwidth used by this job. The number is in bytes/sec,
		the normal suffix rules apply. You can use rate=500k to limit
		reads and writes to 500k each, or you can specify read and
		writes separately. Using rate=1m,500k would limit reads to
		1MB/sec and writes to 500KB/sec. Capping only reads or
		writes can be done with rate=,500k or rate=500k,. The former
		will only limit writes (to 500KB/sec), the latter will only
		limit reads.

rate_min=int	Tell fio to do whatever it can to maintain at least this
		bandwidth. Failing to meet this requirement, will cause
		the job to exit. The same format as rate is used for
		read vs write separation.

rate_iops=int	Cap the bandwidth to this number of IOPS. Basically the same
		as rate, just specified independently of bandwidth. If the
		job is given a block size range instead of a fixed value,
		the smallest block size is used as the metric. The same format
		as rate is used for read vs write separation.

rate_iops_min=int If fio doesn't meet this rate of IO, it will cause
		the job to exit. The same format as rate is used for read vs
		write separation.

rate_process=str	This option controls how fio manages rated IO
		submissions. The default is 'linear', which submits IO in a
		linear fashion with fixed delays between IOs that gets
		adjusted based on IO completion rates. If this is set to
		'poisson', fio will submit IO based on a more real world
		random request flow, known as the Poisson process
		(https://en.wikipedia.org/wiki/Poisson_process). The lambda
		will be 10^6 / IOPS for the given workload.

latency_target=int	If set, fio will attempt to find the max performance
		point that the given workload will run at while maintaining a
		latency below this target. The values is given in microseconds.
		See latency_window and latency_percentile

latency_window=int	Used with latency_target to specify the sample window
		that the job is run at varying queue depths to test the
		performance. The value is given in microseconds.

latency_percentile=float	The percentage of IOs that must fall within the
		criteria specified by latency_target and latency_window. If not
		set, this defaults to 100.0, meaning that all IOs must be equal
		or below to the value set by latency_target.

max_latency=int	If set, fio will exit the job if it exceeds this maximum
		latency. It will exit with an ETIME error.

rate_cycle=int	Average bandwidth for 'rate' and 'rate_min' over this number
		of milliseconds.

cpumask=int	Set the CPU affinity of this job. The parameter given is a
		bitmask of allowed CPU's the job may run on. So if you want
		the allowed CPUs to be 1 and 5, you would pass the decimal
		value of (1 << 1 | 1 << 5), or 34. See man
		sched_setaffinity(2). This may not work on all supported
		operating systems or kernel versions. This option doesn't
		work well for a higher CPU count than what you can store in
		an integer mask, so it can only control cpus 1-32. For
		boxes with larger CPU counts, use cpus_allowed.

cpus_allowed=str Controls the same options as cpumask, but it allows a text
		setting of the permitted CPUs instead. So to use CPUs 1 and
		5, you would specify cpus_allowed=1,5. This options also
		allows a range of CPUs. Say you wanted a binding to CPUs
		1, 5, and 8-15, you would set cpus_allowed=1,5,8-15.

cpus_allowed_policy=str Set the policy of how fio distributes the CPUs
		specified by cpus_allowed or cpumask. Two policies are
		supported:

		shared	All jobs will share the CPU set specified.
		split	Each job will get a unique CPU from the CPU set.

		'shared' is the default behaviour, if the option isn't
		specified. If split is specified, then fio will will assign
		one cpu per job. If not enough CPUs are given for the jobs
		listed, then fio will roundrobin the CPUs in the set.

numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The
		arguments allow comma delimited list of cpu numbers,
		A-B ranges, or 'all'. Note, to enable numa options support,
		fio must be built on a system with libnuma-dev(el) installed.

numa_mem_policy=str Set this job's memory policy and corresponding NUMA
		nodes. Format of the argements:
			<mode>[:<nodelist>]
		`mode' is one of the following memory policy:
			default, prefer, bind, interleave, local
		For `default' and `local' memory policy, no node is
		needed to be specified.
		For `prefer', only one node is allowed.
		For `bind' and `interleave', it allow comma delimited
		list of numbers, A-B ranges, or 'all'.

startdelay=time	Start this job the specified number of seconds after fio
		has started. Only useful if the job file contains several
		jobs, and you want to delay starting some jobs to a certain
		time.

runtime=time	Tell fio to terminate processing after the specified number
		of seconds. It can be quite hard to determine for how long
		a specified job will run, so this parameter is handy to
		cap the total runtime to a given time.

time_based	If set, fio will run for the duration of the runtime
		specified even if the file(s) are completely read or
		written. It will simply loop over the same workload
		as many times as the runtime allows.

ramp_time=time	If set, fio will run the specified workload for this amount
		of time before logging any performance numbers. Useful for
		letting performance settle before logging results, thus
		minimizing the runtime required for stable results. Note
		that the ramp_time is considered lead in time for a job,
		thus it will increase the total runtime if a special timeout
		or runtime is specified.

invalidate=bool	Invalidate the buffer/page cache parts for this file prior
		to starting io. Defaults to true.

sync=bool	Use sync io for buffered writes. For the majority of the
		io engines, this means using O_SYNC.

iomem=str
mem=str		Fio can use various types of memory as the io unit buffer.
		The allowed values are:

			malloc	Use memory from malloc(3) as the buffers.

			shm	Use shared memory as the buffers. Allocated
				through shmget(2).

			shmhuge	Same as shm, but use huge pages as backing.

			mmap	Use mmap to allocate buffers. May either be
				anonymous memory, or can be file backed if
				a filename is given after the option. The
				format is mem=mmap:/path/to/file.

			mmaphuge Use a memory mapped huge file as the buffer
				backing. Append filename after mmaphuge, ala
				mem=mmaphuge:/hugetlbfs/file

			mmapshared	Same as mmap, but use a MMAP_SHARED
				mapping.

		The area allocated is a function of the maximum allowed
		bs size for the job, multiplied by the io depth given. Note
		that for shmhuge and mmaphuge to work, the system must have
		free huge pages allocated. This can normally be checked
		and set by reading/writing /proc/sys/vm/nr_hugepages on a
		Linux system. Fio assumes a huge page is 4MB in size. So
		to calculate the number of huge pages you need for a given
		job file, add up the io depth of all jobs (normally one unless
		iodepth= is used) and multiply by the maximum bs set. Then
		divide that number by the huge page size. You can see the
		size of the huge pages in /proc/meminfo. If no huge pages
		are allocated by having a non-zero number in nr_hugepages,
		using mmaphuge or shmhuge will fail. Also see hugepage-size.

		mmaphuge also needs to have hugetlbfs mounted and the file
		location should point there. So if it's mounted in /huge,
		you would use mem=mmaphuge:/huge/somefile.

iomem_align=int	This indiciates the memory alignment of the IO memory buffers.
		Note that the given alignment is applied to the first IO unit
		buffer, if using iodepth the alignment of the following buffers
		are given by the bs used. In other words, if using a bs that is
		a multiple of the page sized in the system, all buffers will
		be aligned to this value. If using a bs that is not page
		aligned, the alignment of subsequent IO memory buffers is the
		sum of the iomem_align and bs used.

hugepage-size=int
		Defines the size of a huge page. Must at least be equal
		to the system setting, see /proc/meminfo. Defaults to 4MB.
		Should probably always be a multiple of megabytes, so using
		hugepage-size=Xm is the preferred way to set this to avoid
		setting a non-pow-2 bad value.

exitall		When one job finishes, terminate the rest. The default is
		to wait for each job to finish, sometimes that is not the
		desired action.

exitall_on_error	When one job finishes in error, terminate the rest. The
		default is to wait for each job to finish.

bwavgtime=int	Average the calculated bandwidth over the given time. Value
		is specified in milliseconds.

iopsavgtime=int	Average the calculated IOPS over the given time. Value
		is specified in milliseconds.

create_serialize=bool	If true, serialize the file creating for the jobs.
			This may be handy to avoid interleaving of data
			files, which may greatly depend on the filesystem
			used and even the number of processors in the system.

create_fsync=bool	fsync the data file after creation. This is the
			default.

create_on_open=bool	Don't pre-setup the files for IO, just create open()
			when it's time to do IO to that file.

create_only=bool	If true, fio will only run the setup phase of the job.
			If files need to be laid out or updated on disk, only
			that will be done. The actual job contents are not
			executed.

allow_file_create=bool	If true, fio is permitted to create files as part
		of its workload. This is the default behavior. If this
		option is false, then fio will error out if the files it
		needs to use don't already exist. Default: true.

allow_mounted_write=bool	If this isn't set, fio will abort jobs that
		are destructive (eg that write) to what appears to be a
		mounted device or partition. This should help catch creating
		inadvertently destructive tests, not realizing that the test
		will destroy data on the mounted file system. Default: false.

pre_read=bool	If this is given, files will be pre-read into memory before
		starting the given IO operation. This will also clear
		the 'invalidate' flag, since it is pointless to pre-read
		and then drop the cache. This will only work for IO engines
		that are seekable, since they allow you to read the same data
		multiple times. Thus it will not work on eg network or splice
		IO.

unlink=bool	Unlink the job files when done. Not the default, as repeated
		runs of that job would then waste time recreating the file
		set again and again.

loops=int	Run the specified number of iterations of this job. Used
		to repeat the same workload a given number of times. Defaults
		to 1.

verify_only	Do not perform specified workload---only verify data still
		matches previous invocation of this workload. This option
		allows one to check data multiple times at a later date
		without overwriting it. This option makes sense only for
		workloads that write data, and does not support workloads
		with the time_based option set.

do_verify=bool	Run the verify phase after a write phase. Only makes sense if
		verify is set. Defaults to 1.

verify=str	If writing to a file, fio can verify the file contents
		after each iteration of the job. Each verification method also implies
		verification of special header, which is written to the beginning of
		each block. This header also includes meta information, like offset
		of the block, block number, timestamp when block was written, etc.
		verify=str can be combined with verify_pattern=str option.
		The allowed values are:

			md5	Use an md5 sum of the data area and store
				it in the header of each block.

			crc64	Use an experimental crc64 sum of the data
				area and store it in the header of each
				block.

			crc32c	Use a crc32c sum of the data area and store
				it in the header of each block.

			crc32c-intel Use hardware assisted crc32c calcuation
				provided on SSE4.2 enabled processors. Falls
				back to regular software crc32c, if not
				supported by the system.

			crc32	Use a crc32 sum of the data area and store
				it in the header of each block.

			crc16	Use a crc16 sum of the data area and store
				it in the header of each block.

			crc7	Use a crc7 sum of the data area and store
				it in the header of each block.

			xxhash	Use xxhash as the checksum function. Generally
				the fastest software checksum that fio
				supports.

			sha512	Use sha512 as the checksum function.

			sha256	Use sha256 as the checksum function.

			sha1	Use optimized sha1 as the checksum function.

			meta	This option is deprecated, since now meta information is
				included in generic verification header and meta verification
				happens by default. For detailed information see the description
				of the verify=str setting. This option is kept because of
				compatibility's sake with old configurations. Do not use it.

			pattern	Verify a strict pattern. Normally fio includes
				a header with some basic information and
				checksumming, but if this option is set, only
				the specific pattern set with 'verify_pattern'
				is verified.

			null	Only pretend to verify. Useful for testing
				internals with ioengine=null, not for much
				else.

		This option can be used for repeated burn-in tests of a
		system to make sure that the written data is also
		correctly read back. If the data direction given is
		a read or random read, fio will assume that it should
		verify a previously written file. If the data direction
		includes any form of write, the verify will be of the
		newly written data.

verifysort=bool	If set, fio will sort written verify blocks when it deems
		it faster to read them back in a sorted manner. This is
		often the case when overwriting an existing file, since
		the blocks are already laid out in the file system. You
		can ignore this option unless doing huge amounts of really
		fast IO where the red-black tree sorting CPU time becomes
		significant.

verify_offset=int	Swap the verification header with data somewhere else
			in the block before writing. Its swapped back before
			verifying.

verify_interval=int	Write the verification header at a finer granularity
			than the blocksize. It will be written for chunks the
			size of header_interval. blocksize should divide this
			evenly.

verify_pattern=str	If set, fio will fill the io buffers with this
		pattern. Fio defaults to filling with totally random
		bytes, but sometimes it's interesting to fill with a known
		pattern for io verification purposes. Depending on the
		width of the pattern, fio will fill 1/2/3/4 bytes of the
		buffer at the time(it can be either a decimal or a hex number).
		The verify_pattern if larger than a 32-bit quantity has to
		be a hex number that starts with either "0x" or "0X". Use
		with verify=str. Also, verify_pattern supports %o format,
		which means that for each block offset will be written and
		then verifyied back, e.g.:

		verify_pattern=%o

		Or use combination of everything:
		verify_pattern=0xff%o"abcd"-12

verify_fatal=bool	Normally fio will keep checking the entire contents
		before quitting on a block verification failure. If this
		option is set, fio will exit the job on the first observed
		failure.

verify_dump=bool	If set, dump the contents of both the original data
		block and the data block we read off disk to files. This
		allows later analysis to inspect just what kind of data
		corruption occurred. Off by default.

verify_async=int	Fio will normally verify IO inline from the submitting
		thread. This option takes an integer describing how many
		async offload threads to create for IO verification instead,
		causing fio to offload the duty of verifying IO contents
		to one or more separate threads. If using this offload
		option, even sync IO engines can benefit from using an
		iodepth setting higher than 1, as it allows them to have
		IO in flight while verifies are running.

verify_async_cpus=str	Tell fio to set the given CPU affinity on the
		async IO verification threads. See cpus_allowed for the
		format used.

verify_backlog=int	Fio will normally verify the written contents of a
		job that utilizes verify once that job has completed. In
		other words, everything is written then everything is read
		back and verified. You may want to verify continually
		instead for a variety of reasons. Fio stores the meta data
		associated with an IO block in memory, so for large
		verify workloads, quite a bit of memory would be used up
		holding this meta data. If this option is enabled, fio
		will write only N blocks before verifying these blocks.

verify_backlog_batch=int	Control how many blocks fio will verify
		if verify_backlog is set. If not set, will default to
		the value of verify_backlog (meaning the entire queue
		is read back and verified).  If verify_backlog_batch is
		less than verify_backlog then not all blocks will be verified,
		if verify_backlog_batch is larger than verify_backlog, some
		blocks will be verified more than once.

verify_state_save=bool	When a job exits during the write phase of a verify
		workload, save its current state. This allows fio to replay
		up until that point, if the verify state is loaded for the
		verify read phase. The format of the filename is, roughly,
		<type>-<jobname>-<jobindex>-verify.state. <type> is "local"
		for a local run, "sock" for a client/server socket connection,
		and "ip" (192.168.0.1, for instance) for a networked
		client/server connection.

verify_state_load=bool	If a verify termination trigger was used, fio stores
		the current write state of each thread. This can be used at
		verification time so that fio knows how far it should verify.
		Without this information, fio will run a full verification
		pass, according to the settings in the job file used.

stonewall
wait_for_previous Wait for preceding jobs in the job file to exit, before
		starting this one. Can be used to insert serialization
		points in the job file. A stone wall also implies starting
		a new reporting group.

new_group	Start a new reporting group. See: group_reporting.

numjobs=int	Create the specified number of clones of this job. May be
		used to setup a larger number of threads/processes doing
		the same thing. Each thread is reported separately; to see
		statistics for all clones as a whole, use group_reporting in
		conjunction with new_group.

group_reporting	It may sometimes be interesting to display statistics for
		groups of jobs as a whole instead of for each individual job.
		This is especially true if 'numjobs' is used; looking at
		individual thread/process output quickly becomes unwieldy.
		To see the final report per-group instead of per-job, use
		'group_reporting'. Jobs in a file will be part of the same
		reporting group, unless if separated by a stonewall, or by
		using 'new_group'.

thread		fio defaults to forking jobs, however if this option is
		given, fio will use pthread_create(3) to create threads
		instead.

zonesize=int	Divide a file into zones of the specified size. See zoneskip.

zoneskip=int	Skip the specified number of bytes when zonesize data has
		been read. The two zone options can be used to only do
		io on zones of a file.

write_iolog=str	Write the issued io patterns to the specified file. See
		read_iolog.  Specify a separate file for each job, otherwise
		the iologs will be interspersed and the file may be corrupt.

read_iolog=str	Open an iolog with the specified file name and replay the
		io patterns it contains. This can be used to store a
		workload and replay it sometime later. The iolog given
		may also be a blktrace binary file, which allows fio
		to replay a workload captured by blktrace. See blktrace
		for how to capture such logging data. For blktrace replay,
		the file needs to be turned into a blkparse binary data
		file first (blkparse <device> -o /dev/null -d file_for_fio.bin).

replay_no_stall=int When replaying I/O with read_iolog the default behavior
		is to attempt to respect the time stamps within the log and
		replay them with the appropriate delay between IOPS.  By
		setting this variable fio will not respect the timestamps and
		attempt to replay them as fast as possible while still
		respecting ordering.  The result is the same I/O pattern to a
		given device, but different timings.

replay_redirect=str While replaying I/O patterns using read_iolog the
		default behavior is to replay the IOPS onto the major/minor
		device that each IOP was recorded from.  This is sometimes
		undesirable because on a different machine those major/minor
		numbers can map to a different device.  Changing hardware on
		the same system can also result in a different major/minor
		mapping.  Replay_redirect causes all IOPS to be replayed onto
		the single specified device regardless of the device it was
		recorded from. i.e. replay_redirect=/dev/sdc would cause all
		IO in the blktrace to be replayed onto /dev/sdc.  This means
		multiple devices will be replayed onto a single, if the trace
		contains multiple devices.  If you want multiple devices to be
		replayed concurrently to multiple redirected devices you must
		blkparse your trace into separate traces and replay them with
		independent fio invocations.  Unfortuantely this also breaks
		the strict time ordering between multiple device accesses.

replay_align=int	Force alignment of IO offsets and lengths in a trace
		to this power of 2 value.

replay_scale=int	Scale sector offsets down by this factor when
		replaying traces.

per_job_logs=bool	If set, this generates bw/clat/iops log with per
		file private filenames. If not set, jobs with identical names
		will share the log filename. Default: true.

write_bw_log=str If given, write a bandwidth log of the jobs in this job
		file. Can be used to store data of the bandwidth of the
		jobs in their lifetime. The included fio_generate_plots
		script uses gnuplot to turn these text files into nice
		graphs. See write_lat_log for behaviour of given
		filename. For this option, the suffix is _bw.x.log, where
		x is the index of the job (1..N, where N is the number of
		jobs). If 'per_job_logs' is false, then the filename will not
		include the job index.

write_lat_log=str Same as write_bw_log, except that this option stores io
		submission, completion, and total latencies instead. If no
		filename is given with this option, the default filename of
		"jobname_type.log" is used. Even if the filename is given,
		fio will still append the type of log. So if one specifies

		write_lat_log=foo

		The actual log names will be foo_slat.x.log, foo_clat.x.log,
		and foo_lat.x.log, where x is the index of the job (1..N,
		where N is the number of jobs). This helps fio_generate_plot
		fine the logs automatically. If 'per_job_logs' is false, then
		the filename will not include the job index.


write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is
		given with this option, the default filename of
		"jobname_type.x.log" is used,where x is the index of the job
		(1..N, where N is the number of jobs). Even if the filename
		is given, fio will still append the type of log. If
		'per_job_logs' is false, then the filename will not include
		the job index.

log_avg_msec=int By default, fio will log an entry in the iops, latency,
		or bw log for every IO that completes. When writing to the
		disk log, that can quickly grow to a very large size. Setting
		this option makes fio average the each log entry over the
		specified period of time, reducing the resolution of the log.
		See log_max as well. Defaults to 0, logging all entries.

log_max=bool	If log_avg_msec is set, fio logs the average over that window.
		If you instead want to log the maximum value, set this option
		to 1. Defaults to 0, meaning that averaged values are logged.
.
log_offset=int	If this is set, the iolog options will include the byte
		offset for the IO entry as well as the other data values.

log_compression=int	If this is set, fio will compress the IO logs as
		it goes, to keep the memory footprint lower. When a log
		reaches the specified size, that chunk is removed and
		compressed in the background. Given that IO logs are
		fairly highly compressible, this yields a nice memory
		savings for longer runs. The downside is that the
		compression will consume some background CPU cycles, so
		it may impact the run. This, however, is also true if
		the logging ends up consuming most of the system memory.
		So pick your poison. The IO logs are saved normally at the
		end of a run, by decompressing the chunks and storing them
		in the specified log file. This feature depends on the
		availability of zlib.

log_compression_cpus=str	Define the set of CPUs that are allowed to
		handle online log compression for the IO jobs. This can
		provide better isolation between performance sensitive jobs,
		and background compression work.

log_store_compressed=bool	If set, fio will store the log files in a
		compressed format. They can be decompressed with fio, using
		the --inflate-log command line parameter. The files will be
		stored with a .fz suffix.

block_error_percentiles=bool	If set, record errors in trim block-sized
		units from writes and trims and output a histogram of
		how many trims it took to get to errors, and what kind
		of error was encountered.

lockmem=int	Pin down the specified amount of memory with mlock(2). Can
		potentially be used instead of removing memory or booting
		with less memory to simulate a smaller amount of memory.
		The amount specified is per worker.

exec_prerun=str	Before running this job, issue the command specified
		through system(3). Output is redirected in a file called
		jobname.prerun.txt.

exec_postrun=str After the job completes, issue the command specified
		 though system(3). Output is redirected in a file called
		 jobname.postrun.txt.

ioscheduler=str	Attempt to switch the device hosting the file to the specified
		io scheduler before running.

disk_util=bool	Generate disk utilization statistics, if the platform
		supports it. Defaults to on.

disable_lat=bool Disable measurements of total latency numbers. Useful
		only for cutting back the number of calls to gettimeofday,
		as that does impact performance at really high IOPS rates.
		Note that to really get rid of a large amount of these
		calls, this option must be used with disable_slat and
		disable_bw as well.

disable_clat=bool Disable measurements of completion latency numbers. See
		disable_lat.

disable_slat=bool Disable measurements of submission latency numbers. See
		disable_slat.

disable_bw=bool	Disable measurements of throughput/bandwidth numbers. See
		disable_lat.

clat_percentiles=bool Enable the reporting of percentiles of
		 completion latencies.

percentile_list=float_list Overwrite the default list of percentiles
		for completion latencies and the block error histogram.
		Each number is a floating number in the range (0,100],
		and the maximum length of the list is 20. Use ':'
		to separate the numbers, and list the numbers in ascending
		order. For example, --percentile_list=99.5:99.9 will cause
		fio to report the values of completion latency below which
		99.5% and 99.9% of the observed latencies fell, respectively.

clocksource=str	Use the given clocksource as the base of timing. The
		supported options are:

			gettimeofday	gettimeofday(2)

			clock_gettime	clock_gettime(2)

			cpu		Internal CPU clock source

		cpu is the preferred clocksource if it is reliable, as it
		is very fast (and fio is heavy on time calls). Fio will
		automatically use this clocksource if it's supported and
		considered reliable on the system it is running on, unless
		another clocksource is specifically set. For x86/x86-64 CPUs,
		this means supporting TSC Invariant.

gtod_reduce=bool Enable all of the gettimeofday() reducing options
		(disable_clat, disable_slat, disable_bw) plus reduce
		precision of the timeout somewhat to really shrink
		the gettimeofday() call count. With this option enabled,
		we only do about 0.4% of the gtod() calls we would have
		done if all time keeping was enabled.

gtod_cpu=int	Sometimes it's cheaper to dedicate a single thread of
		execution to just getting the current time. Fio (and
		databases, for instance) are very intensive on gettimeofday()
		calls. With this option, you can set one CPU aside for
		doing nothing but logging current time to a shared memory
		location. Then the other threads/processes that run IO
		workloads need only copy that segment, instead of entering
		the kernel with a gettimeofday() call. The CPU set aside
		for doing these time calls will be excluded from other
		uses. Fio will manually clear it from the CPU mask of other
		jobs.

continue_on_error=str	Normally fio will exit the job on the first observed
		failure. If this option is set, fio will continue the job when
		there is a 'non-fatal error' (EIO or EILSEQ) until the runtime
		is exceeded or the I/O size specified is completed. If this
		option is used, there are two more stats that are appended,
		the total error count and the first error. The error field
		given in the stats is the first error that was hit during the
		run.

		The allowed values are:

			none	Exit on any IO or verify errors.

			read	Continue on read errors, exit on all others.

			write	Continue on write errors, exit on all others.

			io	Continue on any IO error, exit on all others.

			verify	Continue on verify errors, exit on all others.

			all	Continue on all errors.

			0		Backward-compatible alias for 'none'.

			1		Backward-compatible alias for 'all'.

ignore_error=str Sometimes you want to ignore some errors during test
		 in that case you can specify error list for each error type.
		 ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST
		 errors for given error type is separated with ':'. Error
		 may be symbol ('ENOSPC', 'ENOMEM') or integer.
		 Example:
			ignore_error=EAGAIN,ENOSPC:122
		 This option will ignore EAGAIN from READ, and ENOSPC and
		 122(EDQUOT) from WRITE.

error_dump=bool If set dump every error even if it is non fatal, true
		by default. If disabled only fatal error will be dumped

cgroup=str	Add job to this control group. If it doesn't exist, it will
		be created. The system must have a mounted cgroup blkio
		mount point for this to work. If your system doesn't have it
		mounted, you can do so with:

		# mount -t cgroup -o blkio none /cgroup

cgroup_weight=int	Set the weight of the cgroup to this value. See
		the documentation that comes with the kernel, allowed values
		are in the range of 100..1000.

cgroup_nodelete=bool Normally fio will delete the cgroups it has created after
		the job completion. To override this behavior and to leave
		cgroups around after the job completion, set cgroup_nodelete=1.
		This can be useful if one wants to inspect various cgroup
		files after job completion. Default: false

uid=int		Instead of running as the invoking user, set the user ID to
		this value before the thread/process does any work.

gid=int		Set group ID, see uid.

flow_id=int	The ID of the flow. If not specified, it defaults to being a
		global flow. See flow.

flow=int	Weight in token-based flow control. If this value is used, then
		there is a 'flow counter' which is used to regulate the
		proportion of activity between two or more jobs. fio attempts
		to keep this flow counter near zero. The 'flow' parameter
		stands for how much should be added or subtracted to the flow
		counter on each iteration of the main I/O loop. That is, if
		one job has flow=8 and another job has flow=-1, then there
		will be a roughly 1:8 ratio in how much one runs vs the other.

flow_watermark=int	The maximum value that the absolute value of the flow
		counter is allowed to reach before the job must wait for a
		lower value of the counter.

flow_sleep=int	The period of time, in microseconds, to wait after the flow
		watermark has been exceeded before retrying operations

In addition, there are some parameters which are only valid when a specific
ioengine is in use. These are used identically to normal parameters, with the
caveat that when used on the command line, they must come after the ioengine
that defines them is selected.

[libaio] userspace_reap Normally, with the libaio engine in use, fio will use
		the io_getevents system call to reap newly returned events.
		With this flag turned on, the AIO ring will be read directly
		from user-space to reap events. The reaping mode is only
		enabled when polling for a minimum of 0 events (eg when
		iodepth_batch_complete=0).

[psyncv2] hipri		Set RWF_HIPRI on IO, indicating to the kernel that
			it's of higher priority than normal.

[cpu] cpuload=int Attempt to use the specified percentage of CPU cycles.

[cpu] cpuchunks=int Split the load into cycles of the given time. In
		microseconds.

[cpu] exit_on_io_done=bool Detect when IO threads are done, then exit.

[netsplice] hostname=str
[net] hostname=str The host name or IP address to use for TCP or UDP based IO.
		If the job is a TCP listener or UDP reader, the hostname is not
		used and must be omitted unless it is a valid UDP multicast
		address.
[libhdfs] namenode=str The host name or IP address of a HDFS cluster namenode to contact.

[netsplice] port=int
[net] port=int	The TCP or UDP port to bind to or connect to. If this is used
with numjobs to spawn multiple instances of the same job type, then this will
be the starting port number since fio will use a range of ports.
[libhdfs] port=int	the listening port of the HFDS cluster namenode.

[netsplice] interface=str
[net] interface=str  The IP address of the network interface used to send or
		receive UDP multicast

[netsplice] ttl=int
[net] ttl=int	Time-to-live value for outgoing UDP multicast packets.
		Default: 1

[netsplice] nodelay=bool
[net] nodelay=bool	Set TCP_NODELAY on TCP connections.

[netsplice] protocol=str
[netsplice] proto=str
[net] protocol=str
[net] proto=str	The network protocol to use. Accepted values are:

			tcp	Transmission control protocol
			tcpv6	Transmission control protocol V6
			udp	User datagram protocol
			udpv6	User datagram protocol V6
			unix	UNIX domain socket

		When the protocol is TCP or UDP, the port must also be given,
		as well as the hostname if the job is a TCP listener or UDP
		reader. For unix sockets, the normal filename option should be
		used and the port is invalid.

[net] listen	For TCP network connections, tell fio to listen for incoming
		connections rather than initiating an outgoing connection. The
		hostname must be omitted if this option is used.

[net] pingpong	Normaly a network writer will just continue writing data, and
		a network reader will just consume packages. If pingpong=1
		is set, a writer will send its normal payload to the reader,
		then wait for the reader to send the same payload back. This
		allows fio to measure network latencies. The submission
		and completion latencies then measure local time spent
		sending or receiving, and the completion latency measures
		how long it took for the other end to receive and send back.
		For UDP multicast traffic pingpong=1 should only be set for a
		single reader when multiple readers are listening to the same
		address.

[net] window_size	Set the desired socket buffer size for the connection.

[net] mss	Set the TCP maximum segment size (TCP_MAXSEG).

[e4defrag] donorname=str
	        File will be used as a block donor(swap extents between files)
[e4defrag] inplace=int
		Configure donor file blocks allocation strategy
		0(default): Preallocate donor's file on init
		1 	  : allocate space immidietly inside defragment event,
			    and free right after event

[mtd] skip_bad=bool	Skip operations against known bad blocks.

[libhdfs] hdfsdirectory	libhdfs will create chunk in this HDFS directory
[libhdfs] chunck_size	the size of the chunck to use for each file.


6.0 Interpreting the output
---------------------------

fio spits out a lot of output. While running, fio will display the
status of the jobs created. An example of that would be:

Threads: 1: [_r] [24.8% done] [ 13509/  8334 kb/s] [eta 00h:01m:31s]

The characters inside the square brackets denote the current status of
each thread. The possible values (in typical life cycle order) are:

Idle	Run
----    ---
P		Thread setup, but not started.
C		Thread created.
I		Thread initialized, waiting or generating necessary data.
	p	Thread running pre-reading file(s).
	R	Running, doing sequential reads.
	r	Running, doing random reads.
	W	Running, doing sequential writes.
	w	Running, doing random writes.
	M	Running, doing mixed sequential reads/writes.
	m	Running, doing mixed random reads/writes.
	F	Running, currently waiting for fsync()
	f	Running, finishing up (writing IO logs, etc)
	V	Running, doing verification of written data.
E		Thread exited, not reaped by main thread yet.
_		Thread reaped, or
X		Thread reaped, exited with an error.
K		Thread reaped, exited due to signal.

Fio will condense the thread string as not to take up more space on the
command line as is needed. For instance, if you have 10 readers and 10
writers running, the output would look like this:

Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s]

Fio will still maintain the ordering, though. So the above means that jobs
1..10 are readers, and 11..20 are writers.

The other values are fairly self explanatory - number of threads
currently running and doing io, rate of io since last check (read speed
listed first, then write speed), and the estimated completion percentage
and time for the running group. It's impossible to estimate runtime of
the following groups (if any). Note that the string is displayed in order,
so it's possible to tell which of the jobs are currently doing what. The
first character is the first job defined in the job file, and so forth.

When fio is done (or interrupted by ctrl-c), it will show the data for
each thread, group of threads, and disks in that order. For each data
direction, the output looks like:

Client1 (g=0): err= 0:
  write: io=    32MB, bw=   666KB/s, iops=89 , runt= 50320msec
    slat (msec): min=    0, max=  136, avg= 0.03, stdev= 1.92
    clat (msec): min=    0, max=  631, avg=48.50, stdev=86.82
    bw (KB/s) : min=    0, max= 1196, per=51.00%, avg=664.02, stdev=681.68
  cpu        : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17
  IO depths    : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0%
     submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
     issued r/w: total=0/32768, short=0/0
     lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%,
     lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0%

The client number is printed, along with the group id and error of that
thread. Below is the io statistics, here for writes. In the order listed,
they denote:

io=		Number of megabytes io performed
bw=		Average bandwidth rate
iops=           Average IOs performed per second
runt=		The runtime of that thread
	slat=	Submission latency (avg being the average, stdev being the
		standard deviation). This is the time it took to submit
		the io. For sync io, the slat is really the completion
		latency, since queue/complete is one operation there. This
		value can be in milliseconds or microseconds, fio will choose
		the most appropriate base and print that. In the example
		above, milliseconds is the best scale. Note: in --minimal mode
		latencies are always expressed in microseconds.
	clat=	Completion latency. Same names as slat, this denotes the
		time from submission to completion of the io pieces. For
		sync io, clat will usually be equal (or very close) to 0,
		as the time from submit to complete is basically just
		CPU time (io has already been done, see slat explanation).
	bw=	Bandwidth. Same names as the xlat stats, but also includes
		an approximate percentage of total aggregate bandwidth
		this thread received in this group. This last value is
		only really useful if the threads in this group are on the
		same disk, since they are then competing for disk access.
cpu=		CPU usage. User and system time, along with the number
		of context switches this thread went through, usage of
		system and user time, and finally the number of major
		and minor page faults.
IO depths=	The distribution of io depths over the job life time. The
		numbers are divided into powers of 2, so for example the
		16= entries includes depths up to that value but higher
		than the previous entry. In other words, it covers the
		range from 16 to 31.
IO submit=	How many pieces of IO were submitting in a single submit
		call. Each entry denotes that amount and below, until
		the previous entry - eg, 8=100% mean that we submitted
		anywhere in between 5-8 ios per submit call.
IO complete=	Like the above submit number, but for completions instead.
IO issued=	The number of read/write requests issued, and how many
		of them were short.
IO latencies=	The distribution of IO completion latencies. This is the
		time from when IO leaves fio and when it gets completed.
		The numbers follow the same pattern as the IO depths,
		meaning that 2=1.6% means that 1.6% of the IO completed
		within 2 msecs, 20=12.8% means that 12.8% of the IO
		took more than 10 msecs, but less than (or equal to) 20 msecs.

After each client has been listed, the group statistics are printed. They
will look like this:

Run status group 0 (all jobs):
   READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec
  WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec

For each data direction, it prints:

io=		Number of megabytes io performed.
aggrb=		Aggregate bandwidth of threads in this group.
minb=		The minimum average bandwidth a thread saw.
maxb=		The maximum average bandwidth a thread saw.
mint=		The smallest runtime of the threads in that group.
maxt=		The longest runtime of the threads in that group.

And finally, the disk statistics are printed. They will look like this:

Disk stats (read/write):
  sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00%

Each value is printed for both reads and writes, with reads first. The
numbers denote:

ios=		Number of ios performed by all groups.
merge=		Number of merges io the io scheduler.
ticks=		Number of ticks we kept the disk busy.
io_queue=	Total time spent in the disk queue.
util=		The disk utilization. A value of 100% means we kept the disk
		busy constantly, 50% would be a disk idling half of the time.

It is also possible to get fio to dump the current output while it is
running, without terminating the job. To do that, send fio the USR1 signal.
You can also get regularly timed dumps by using the --status-interval
parameter, or by creating a file in /tmp named fio-dump-status. If fio
sees this file, it will unlink it and dump the current output status.


7.0 Terse output
----------------

For scripted usage where you typically want to generate tables or graphs
of the results, fio can output the results in a semicolon separated format.
The format is one long line of values, such as:

2;card0;0;0;7139336;121836;60004;1;10109;27.932460;116.933948;220;126861;3495.446807;1085.368601;226;126864;3523.635629;1089.012448;24063;99944;50.275485%;59818.274627;5540.657370;7155060;122104;60004;1;8338;29.086342;117.839068;388;128077;5032.488518;1234.785715;391;128085;5061.839412;1236.909129;23436;100928;50.287926%;59964.832030;5644.844189;14.595833%;19.394167%;123706;0;7313;0.1%;0.1%;0.1%;0.1%;0.1%;0.1%;100.0%;0.00%;0.00%;0.00%;0.00%;0.00%;0.00%;0.01%;0.02%;0.05%;0.16%;6.04%;40.40%;52.68%;0.64%;0.01%;0.00%;0.01%;0.00%;0.00%;0.00%;0.00%;0.00%
A description of this job goes here.

The job description (if provided) follows on a second line.

To enable terse output, use the --minimal command line option. The first
value is the version of the terse output format. If the output has to
be changed for some reason, this number will be incremented by 1 to
signify that change.

Split up, the format is as follows:

	terse version, fio version, jobname, groupid, error
	READ status:
		Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
		Submission latency: min, max, mean, stdev (usec)
		Completion latency: min, max, mean, stdev (usec)
		Completion latency percentiles: 20 fields (see below)
		Total latency: min, max, mean, stdev (usec)
		Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
	WRITE status:
		Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec)
		Submission latency: min, max, mean, stdev (usec)
		Completion latency: min, max, mean, stdev(usec)
		Completion latency percentiles: 20 fields (see below)
		Total latency: min, max, mean, stdev (usec)
		Bw (KB/s): min, max, aggregate percentage of total, mean, stdev
	CPU usage: user, system, context switches, major faults, minor faults
	IO depths: <=1, 2, 4, 8, 16, 32, >=64
	IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000
	IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000
	Disk utilization: Disk name, Read ios, write ios,
			  Read merges, write merges,
			  Read ticks, write ticks,
			  Time spent in queue, disk utilization percentage
	Additional Info (dependent on continue_on_error, default off): total # errors, first error code

	Additional Info (dependent on description being set): Text description

Completion latency percentiles can be a grouping of up to 20 sets, so
for the terse output fio writes all of them. Each field will look like this:

	1.00%=6112

which is the Xth percentile, and the usec latency associated with it.

For disk utilization, all disks used by fio are shown. So for each disk
there will be a disk utilization section.


8.0 Trace file format
---------------------
There are two trace file format that you can encounter. The older (v1) format
is unsupported since version 1.20-rc3 (March 2008). It will still be described
below in case that you get an old trace and want to understand it.

In any case the trace is a simple text file with a single action per line.


8.1 Trace file format v1
------------------------
Each line represents a single io action in the following format:

rw, offset, length

where rw=0/1 for read/write, and the offset and length entries being in bytes.

This format is not supported in Fio versions => 1.20-rc3.


8.2 Trace file format v2
------------------------
The second version of the trace file format was added in Fio version 1.17.
It allows to access more then one file per trace and has a bigger set of
possible file actions.

The first line of the trace file has to be:

fio version 2 iolog

Following this can be lines in two different formats, which are described below.

The file management format:

filename action

The filename is given as an absolute path. The action can be one of these:

add          Add the given filename to the trace
open         Open the file with the given filename. The filename has to have
             been added with the add action before.
close        Close the file with the given filename. The file has to have been
             opened before.


The file io action format:

filename action offset length

The filename is given as an absolute path, and has to have been added and opened
before it can be used with this format. The offset and length are given in
bytes. The action can be one of these:

wait       Wait for 'offset' microseconds. Everything below 100 is discarded.
	   The time is relative to the previous wait statement.
read       Read 'length' bytes beginning from 'offset'
write      Write 'length' bytes beginning from 'offset'
sync       fsync() the file
datasync   fdatasync() the file
trim       trim the given file from the given 'offset' for 'length' bytes


9.0 CPU idleness profiling
--------------------------
In some cases, we want to understand CPU overhead in a test. For example,
we test patches for the specific goodness of whether they reduce CPU usage.
fio implements a balloon approach to create a thread per CPU that runs at
idle priority, meaning that it only runs when nobody else needs the cpu.
By measuring the amount of work completed by the thread, idleness of each
CPU can be derived accordingly.

An unit work is defined as touching a full page of unsigned characters. Mean
and standard deviation of time to complete an unit work is reported in "unit
work" section. Options can be chosen to report detailed percpu idleness or
overall system idleness by aggregating percpu stats.


10.0 Verification and triggers
------------------------------
Fio is usually run in one of two ways, when data verification is done. The
first is a normal write job of some sort with verify enabled. When the
write phase has completed, fio switches to reads and verifies everything
it wrote. The second model is running just the write phase, and then later
on running the same job (but with reads instead of writes) to repeat the
same IO patterns and verify the contents. Both of these methods depend
on the write phase being completed, as fio otherwise has no idea how much
data was written.

With verification triggers, fio supports dumping the current write state
to local files. Then a subsequent read verify workload can load this state
and know exactly where to stop. This is useful for testing cases where
power is cut to a server in a managed fashion, for instance.

A verification trigger consists of two things:

1) Storing the write state of each job
2) Executing a trigger command

The write state is relatively small, on the order of hundreds of bytes
to single kilobytes. It contains information on the number of completions
done, the last X completions, etc.

A trigger is invoked either through creation ('touch') of a specified
file in the system, or through a timeout setting. If fio is run with
--trigger-file=/tmp/trigger-file, then it will continually check for
the existence of /tmp/trigger-file. When it sees this file, it will
fire off the trigger (thus saving state, and executing the trigger
command).

For client/server runs, there's both a local and remote trigger. If
fio is running as a server backend, it will send the job states back
to the client for safe storage, then execute the remote trigger, if
specified. If a local trigger is specified, the server will still send
back the write state, but the client will then execute the trigger.

10.1 Verification trigger example
---------------------------------
Lets say we want to run a powercut test on the remote machine 'server'.
Our write workload is in write-test.fio. We want to cut power to 'server'
at some point during the run, and we'll run this test from the safety
or our local machine, 'localbox'. On the server, we'll start the fio
backend normally:

server# fio --server

and on the client, we'll fire off the workload:

localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\""

We set /tmp/my-trigger as the trigger file, and we tell fio to execute

echo b > /proc/sysrq-trigger

on the server once it has received the trigger and sent us the write
state. This will work, but it's not _really_ cutting power to the server,
it's merely abruptly rebooting it. If we have a remote way of cutting
power to the server through IPMI or similar, we could do that through
a local trigger command instead. Lets assume we have a script that does
IPMI reboot of a given hostname, ipmi-reboot. On localbox, we could
then have run fio with a local trigger instead:

localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server"

For this case, fio would wait for the server to send us the write state,
then execute 'ipmi-reboot server' when that happened.

10.2 Loading verify state
-------------------------
To load store write state, read verification job file must contain
the verify_state_load option. If that is set, fio will load the previously
stored state. For a local fio run this is done by loading the files directly,
and on a client/server run, the server backend will ask the client to send
the files over and load them from there.