fio2.1.10--HOWTO

扫码查看

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

---------------使用命令行或者job file文件

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 the 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

-------------------job file的格式怎么定义,可以参考example文件夹下的例子,其中有两个关于ssd的fio参考事例

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

4.1 Environment variables

-------------------------还可以使用环境变量作为参数

fio also supports environment variable expansion in job files. Any
substring 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
---------------------保留的关键字

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. 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.

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

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
            one 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.

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.

use_os_rand=bool Fio can either use the random generator supplied by the
OS
            to generator random offsets,
or it can use it's own internal
            generator (based on
Tausworthe). Default is to use the
            internal generator, which is
often of better quality and
            faster.

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.

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).
            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 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_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_limit'
is set to 5G, fio will perform IO within the first
            20G but exit when 5G have been
done.

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.
            The resulting IO buffers will
not be completely zeroed,
            unless scramble_buffers is
also turned off.

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 zeroes. 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.

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

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.

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_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.

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
            the offset + offset_increment
* thread_number, where the
            thread number is a counter
that starts at 0 and is incremented
            for each job. This option is
useful if there are several jobs
            which are intended to operate
on a file in parallel in 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.

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

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.

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. This option
            is mutually exclusive with
verify= if and only if multiple
            blocksizes (via bsrange=) are
used, since fio only tracks
            complete rewrites of
blocks.

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

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.

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.

ratemin=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.

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.

ratecycle=int    Average bandwidth for
'rate' and 'ratemin' 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

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.

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.

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. 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    Write extra information about each io
                       (timestamp, block
number etc.). The block
                       number is verified.
The io sequence number is
                        verified for workloads that write data.
                       See also
verify_pattern.

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=meta.

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.

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.

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.log.

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.log, foo_clat.log,
            and foo_lat.log. This helps
fio_generate_plot fine the logs
            automatically.

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.log"
is used. Even if the filename is given,
            fio will still append the type
of log.

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.
            Defaults to 0.

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. 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).

[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.

[netsplice] port=int
[net] port=int   The TCP or UDP port to
bind to or connect to.

[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.

[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

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.

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, deviation (usec)
            Completion latency: min, max,
mean, deviation (usec)
            Completion latency
percentiles: 20 fields (see below)
            Total latency: min, max, mean,
deviation (usec)
            Bw (KB/s): min, max, aggregate
percentage of total, mean, deviation
      WRITE status:
            Total IO (KB), bandwidth
(KB/sec), IOPS, runtime (msec)
            Submission latency: min, max,
mean, deviation (usec)
            Completion latency: min, max,
mean, deviation (usec)
            Completion latency
percentiles: 20 fields (see below)
            Total latency: min, max, mean,
deviation (usec)
            Bw (KB/s): min, max, aggregate
percentage of total, mean, deviation
      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.
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.

05-02 16:25
查看更多