blob: 0f7909daef68c32e0e4cee1546d4b93a479a3e0a [file] [log] [blame]
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
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 --
; -- 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 --
; -- 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
$ 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 --
include glob-include.fio
include test-include.fio
; -- end job file including.fio --
; -- start job file glob-include.fio --
; -- end job file glob-include.fio --
; -- start job file test-include.fio --
; -- 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 []
4.1 Environment variables
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
As an example, let's look at a sample fio invocation and job file:
$ SIZE=64m NUMJOBS=4 fio jobfile.fio
; -- start job file --
; -- end job file --
This will expand to the following equivalent job file at runtime:
; -- start job file --
; -- end job file --
fio ships with a few example job files, you can also look there for
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:
and get that properly expanded to 8 times the size of memory in the
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
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
float_list A list of floating numbers, separated by a ':' character.
With the above in mind, here follows the complete list of fio job
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
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.
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:
The name of the worker thread or process.
The incremental number of the worker thread or
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
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
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
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"
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
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
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,
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_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_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.
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.
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.
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:
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:
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:
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
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 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_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 "".
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
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
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
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).
The 'filename' option is used to specify host,
port of the hdfs name-node to connect. 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
expects bunch of small files to be created
over HDFS, and engine will randomly pick a
file out of those files based on the offset
generated by fio backend. (see the example
job file to create such files, use rw=write
option). Please note, you might want to set
necessary environment variables to work with
hdfs/libhdfs properly.
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=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
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:
So if you do sync_file_range=wait_before,write:8, fio would
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. 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
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
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
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
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' 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
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.
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
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.
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
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
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
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
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
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
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
verify_offset=int Swap the verification header with data somewhere else
in the block before writing. Its swapped back before
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
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
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" (, 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.
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
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.x.log, where
x is the index of the job (1..N, where N is the number of
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
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.
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.
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.
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_store_compressed=bool If set, and log_compression is also 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.
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
exec_postrun=str After the job completes, issue the command specified
though system(3). Output is redirected in a file called
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_slat=bool Disable measurements of submission latency numbers. See
disable_bw=bool Disable measurements of throughput/bandwidth numbers. See
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
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
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.
errors for given error type is separated with ':'. Error
may be symbol ('ENOSPC', 'ENOMEM') or integer.
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
[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
[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
[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.
[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
[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
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:
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:
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.
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
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.
10.1 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.