| 1 | Table of contents |
| 2 | ----------------- |
| 3 | |
| 4 | 1. Overview |
| 5 | 2. How fio works |
| 6 | 3. Running fio |
| 7 | 4. Job file format |
| 8 | 5. Detailed list of parameters |
| 9 | 6. Normal output |
| 10 | 7. Terse output |
| 11 | 8. Trace file format |
| 12 | 9. CPU idleness profiling |
| 13 | 10. Verification and triggers |
| 14 | |
| 15 | 1.0 Overview and history |
| 16 | ------------------------ |
| 17 | fio was originally written to save me the hassle of writing special test |
| 18 | case programs when I wanted to test a specific workload, either for |
| 19 | performance reasons or to find/reproduce a bug. The process of writing |
| 20 | such a test app can be tiresome, especially if you have to do it often. |
| 21 | Hence I needed a tool that would be able to simulate a given io workload |
| 22 | without resorting to writing a tailored test case again and again. |
| 23 | |
| 24 | A test work load is difficult to define, though. There can be any number |
| 25 | of processes or threads involved, and they can each be using their own |
| 26 | way of generating io. You could have someone dirtying large amounts of |
| 27 | memory in an memory mapped file, or maybe several threads issuing |
| 28 | reads using asynchronous io. fio needed to be flexible enough to |
| 29 | simulate both of these cases, and many more. |
| 30 | |
| 31 | 2.0 How fio works |
| 32 | ----------------- |
| 33 | The first step in getting fio to simulate a desired io workload, is |
| 34 | writing a job file describing that specific setup. A job file may contain |
| 35 | any number of threads and/or files - the typical contents of the job file |
| 36 | is a global section defining shared parameters, and one or more job |
| 37 | sections describing the jobs involved. When run, fio parses this file |
| 38 | and sets everything up as described. If we break down a job from top to |
| 39 | bottom, it contains the following basic parameters: |
| 40 | |
| 41 | IO type Defines the io pattern issued to the file(s). |
| 42 | We may only be reading sequentially from this |
| 43 | file(s), or we may be writing randomly. Or even |
| 44 | mixing reads and writes, sequentially or randomly. |
| 45 | |
| 46 | Block size In how large chunks are we issuing io? This may be |
| 47 | a single value, or it may describe a range of |
| 48 | block sizes. |
| 49 | |
| 50 | IO size How much data are we going to be reading/writing. |
| 51 | |
| 52 | IO engine How do we issue io? We could be memory mapping the |
| 53 | file, we could be using regular read/write, we |
| 54 | could be using splice, async io, syslet, or even |
| 55 | SG (SCSI generic sg). |
| 56 | |
| 57 | IO depth If the io engine is async, how large a queuing |
| 58 | depth do we want to maintain? |
| 59 | |
| 60 | IO type Should we be doing buffered io, or direct/raw io? |
| 61 | |
| 62 | Num files How many files are we spreading the workload over. |
| 63 | |
| 64 | Num threads How many threads or processes should we spread |
| 65 | this workload over. |
| 66 | |
| 67 | The above are the basic parameters defined for a workload, in addition |
| 68 | there's a multitude of parameters that modify other aspects of how this |
| 69 | job behaves. |
| 70 | |
| 71 | |
| 72 | 3.0 Running fio |
| 73 | --------------- |
| 74 | See the README file for command line parameters, there are only a few |
| 75 | of them. |
| 76 | |
| 77 | Running fio is normally the easiest part - you just give it the job file |
| 78 | (or job files) as parameters: |
| 79 | |
| 80 | $ fio job_file |
| 81 | |
| 82 | and it will start doing what the job_file tells it to do. You can give |
| 83 | more than one job file on the command line, fio will serialize the running |
| 84 | of those files. Internally that is the same as using the 'stonewall' |
| 85 | parameter described in the parameter section. |
| 86 | |
| 87 | If the job file contains only one job, you may as well just give the |
| 88 | parameters on the command line. The command line parameters are identical |
| 89 | to the job parameters, with a few extra that control global parameters |
| 90 | (see README). For example, for the job file parameter iodepth=2, the |
| 91 | mirror command line option would be --iodepth 2 or --iodepth=2. You can |
| 92 | also use the command line for giving more than one job entry. For each |
| 93 | --name option that fio sees, it will start a new job with that name. |
| 94 | Command line entries following a --name entry will apply to that job, |
| 95 | until there are no more entries or a new --name entry is seen. This is |
| 96 | similar to the job file options, where each option applies to the current |
| 97 | job until a new [] job entry is seen. |
| 98 | |
| 99 | fio does not need to run as root, except if the files or devices specified |
| 100 | in the job section requires that. Some other options may also be restricted, |
| 101 | such as memory locking, io scheduler switching, and decreasing the nice value. |
| 102 | |
| 103 | |
| 104 | 4.0 Job file format |
| 105 | ------------------- |
| 106 | As previously described, fio accepts one or more job files describing |
| 107 | what it is supposed to do. The job file format is the classic ini file, |
| 108 | where the names enclosed in [] brackets define the job name. You are free |
| 109 | to use any ascii name you want, except 'global' which has special meaning. |
| 110 | A global section sets defaults for the jobs described in that file. A job |
| 111 | may override a global section parameter, and a job file may even have |
| 112 | several global sections if so desired. A job is only affected by a global |
| 113 | section residing above it. If the first character in a line is a ';' or a |
| 114 | '#', the entire line is discarded as a comment. |
| 115 | |
| 116 | So let's look at a really simple job file that defines two processes, each |
| 117 | randomly reading from a 128MB file. |
| 118 | |
| 119 | ; -- start job file -- |
| 120 | [global] |
| 121 | rw=randread |
| 122 | size=128m |
| 123 | |
| 124 | [job1] |
| 125 | |
| 126 | [job2] |
| 127 | |
| 128 | ; -- end job file -- |
| 129 | |
| 130 | As you can see, the job file sections themselves are empty as all the |
| 131 | described parameters are shared. As no filename= option is given, fio |
| 132 | makes up a filename for each of the jobs as it sees fit. On the command |
| 133 | line, this job would look as follows: |
| 134 | |
| 135 | $ fio --name=global --rw=randread --size=128m --name=job1 --name=job2 |
| 136 | |
| 137 | |
| 138 | Let's look at an example that has a number of processes writing randomly |
| 139 | to files. |
| 140 | |
| 141 | ; -- start job file -- |
| 142 | [random-writers] |
| 143 | ioengine=libaio |
| 144 | iodepth=4 |
| 145 | rw=randwrite |
| 146 | bs=32k |
| 147 | direct=0 |
| 148 | size=64m |
| 149 | numjobs=4 |
| 150 | |
| 151 | ; -- end job file -- |
| 152 | |
| 153 | Here we have no global section, as we only have one job defined anyway. |
| 154 | We want to use async io here, with a depth of 4 for each file. We also |
| 155 | increased the buffer size used to 32KB and define numjobs to 4 to |
| 156 | fork 4 identical jobs. The result is 4 processes each randomly writing |
| 157 | to their own 64MB file. Instead of using the above job file, you could |
| 158 | have given the parameters on the command line. For this case, you would |
| 159 | specify: |
| 160 | |
| 161 | $ fio --name=random-writers --ioengine=libaio --iodepth=4 --rw=randwrite --bs=32k --direct=0 --size=64m --numjobs=4 |
| 162 | |
| 163 | When fio is utilized as a basis of any reasonably large test suite, it might be |
| 164 | desirable to share a set of standardized settings across multiple job files. |
| 165 | Instead of copy/pasting such settings, any section may pull in an external |
| 166 | .fio file with 'include filename' directive, as in the following example: |
| 167 | |
| 168 | ; -- start job file including.fio -- |
| 169 | [global] |
| 170 | filename=/tmp/test |
| 171 | filesize=1m |
| 172 | include glob-include.fio |
| 173 | |
| 174 | [test] |
| 175 | rw=randread |
| 176 | bs=4k |
| 177 | time_based=1 |
| 178 | runtime=10 |
| 179 | include test-include.fio |
| 180 | ; -- end job file including.fio -- |
| 181 | |
| 182 | ; -- start job file glob-include.fio -- |
| 183 | thread=1 |
| 184 | group_reporting=1 |
| 185 | ; -- end job file glob-include.fio -- |
| 186 | |
| 187 | ; -- start job file test-include.fio -- |
| 188 | ioengine=libaio |
| 189 | iodepth=4 |
| 190 | ; -- end job file test-include.fio -- |
| 191 | |
| 192 | Settings pulled into a section apply to that section only (except global |
| 193 | section). Include directives may be nested in that any included file may |
| 194 | contain further include directive(s). Include files may not contain [] |
| 195 | sections. |
| 196 | |
| 197 | |
| 198 | 4.1 Environment variables |
| 199 | ------------------------- |
| 200 | |
| 201 | fio also supports environment variable expansion in job files. Any |
| 202 | sub-string of the form "${VARNAME}" as part of an option value (in other |
| 203 | words, on the right of the `='), will be expanded to the value of the |
| 204 | environment variable called VARNAME. If no such environment variable |
| 205 | is defined, or VARNAME is the empty string, the empty string will be |
| 206 | substituted. |
| 207 | |
| 208 | As an example, let's look at a sample fio invocation and job file: |
| 209 | |
| 210 | $ SIZE=64m NUMJOBS=4 fio jobfile.fio |
| 211 | |
| 212 | ; -- start job file -- |
| 213 | [random-writers] |
| 214 | rw=randwrite |
| 215 | size=${SIZE} |
| 216 | numjobs=${NUMJOBS} |
| 217 | ; -- end job file -- |
| 218 | |
| 219 | This will expand to the following equivalent job file at runtime: |
| 220 | |
| 221 | ; -- start job file -- |
| 222 | [random-writers] |
| 223 | rw=randwrite |
| 224 | size=64m |
| 225 | numjobs=4 |
| 226 | ; -- end job file -- |
| 227 | |
| 228 | fio ships with a few example job files, you can also look there for |
| 229 | inspiration. |
| 230 | |
| 231 | 4.2 Reserved keywords |
| 232 | --------------------- |
| 233 | |
| 234 | Additionally, fio has a set of reserved keywords that will be replaced |
| 235 | internally with the appropriate value. Those keywords are: |
| 236 | |
| 237 | $pagesize The architecture page size of the running system |
| 238 | $mb_memory Megabytes of total memory in the system |
| 239 | $ncpus Number of online available CPUs |
| 240 | |
| 241 | These can be used on the command line or in the job file, and will be |
| 242 | automatically substituted with the current system values when the job |
| 243 | is run. Simple math is also supported on these keywords, so you can |
| 244 | perform actions like: |
| 245 | |
| 246 | size=8*$mb_memory |
| 247 | |
| 248 | and get that properly expanded to 8 times the size of memory in the |
| 249 | machine. |
| 250 | |
| 251 | |
| 252 | 5.0 Detailed list of parameters |
| 253 | ------------------------------- |
| 254 | |
| 255 | This section describes in details each parameter associated with a job. |
| 256 | Some parameters take an option of a given type, such as an integer or |
| 257 | a string. Anywhere a numeric value is required, an arithmetic expression |
| 258 | may be used, provided it is surrounded by parentheses. Supported operators |
| 259 | are: |
| 260 | |
| 261 | addition (+) |
| 262 | subtraction (-) |
| 263 | multiplication (*) |
| 264 | division (/) |
| 265 | modulus (%) |
| 266 | exponentiation (^) |
| 267 | |
| 268 | For time values in expressions, units are microseconds by default. This is |
| 269 | different than for time values not in expressions (not enclosed in |
| 270 | parentheses). The following types are used: |
| 271 | |
| 272 | str String. This is a sequence of alpha characters. |
| 273 | time Integer with possible time suffix. In seconds unless otherwise |
| 274 | specified, use eg 10m for 10 minutes. Accepts s/m/h for seconds, |
| 275 | minutes, and hours, and accepts 'ms' (or 'msec') for milliseconds, |
| 276 | and 'us' (or 'usec') for microseconds. |
| 277 | int SI integer. A whole number value, which may contain a suffix |
| 278 | describing the base of the number. Accepted suffixes are k/m/g/t/p, |
| 279 | meaning kilo, mega, giga, tera, and peta. The suffix is not case |
| 280 | sensitive, and you may also include trailing 'b' (eg 'kb' is the same |
| 281 | as 'k'). So if you want to specify 4096, you could either write |
| 282 | out '4096' or just give 4k. The suffixes signify base 2 values, so |
| 283 | 1024 is 1k and 1024k is 1m and so on, unless the suffix is explicitly |
| 284 | set to a base 10 value using 'kib', 'mib', 'gib', etc. If that is the |
| 285 | case, then 1000 is used as the multiplier. This can be handy for |
| 286 | disks, since manufacturers generally use base 10 values when listing |
| 287 | the capacity of a drive. If the option accepts an upper and lower |
| 288 | range, use a colon ':' or minus '-' to separate such values. May also |
| 289 | include a prefix to indicate numbers base. If 0x is used, the number |
| 290 | is assumed to be hexadecimal. See irange. |
| 291 | bool Boolean. Usually parsed as an integer, however only defined for |
| 292 | true and false (1 and 0). |
| 293 | irange Integer range with suffix. Allows value range to be given, such |
| 294 | as 1024-4096. A colon may also be used as the separator, eg |
| 295 | 1k:4k. If the option allows two sets of ranges, they can be |
| 296 | specified with a ',' or '/' delimiter: 1k-4k/8k-32k. Also see |
| 297 | int. |
| 298 | float_list A list of floating numbers, separated by a ':' character. |
| 299 | |
| 300 | With the above in mind, here follows the complete list of fio job |
| 301 | parameters. |
| 302 | |
| 303 | name=str ASCII name of the job. This may be used to override the |
| 304 | name printed by fio for this job. Otherwise the job |
| 305 | name is used. On the command line this parameter has the |
| 306 | special purpose of also signaling the start of a new |
| 307 | job. |
| 308 | |
| 309 | wait_for=str Specifies the name of the already defined job to wait |
| 310 | for. Single waitee name only may be specified. If set, the job |
| 311 | won't be started until all workers of the waitee job are done. |
| 312 | |
| 313 | Wait_for operates on the job name basis, so there are a few |
| 314 | limitations. First, the waitee must be defined prior to the |
| 315 | waiter job (meaning no forward references). Second, if a job |
| 316 | is being referenced as a waitee, it must have a unique name |
| 317 | (no duplicate waitees). |
| 318 | |
| 319 | description=str Text description of the job. Doesn't do anything except |
| 320 | dump this text description when this job is run. It's |
| 321 | not parsed. |
| 322 | |
| 323 | directory=str Prefix filenames with this directory. Used to place files |
| 324 | in a different location than "./". See the 'filename' option |
| 325 | for escaping certain characters. |
| 326 | |
| 327 | filename=str Fio normally makes up a filename based on the job name, |
| 328 | thread number, and file number. If you want to share |
| 329 | files between threads in a job or several jobs, specify |
| 330 | a filename for each of them to override the default. If |
| 331 | the ioengine used is 'net', the filename is the host, port, |
| 332 | and protocol to use in the format of =host,port,protocol. |
| 333 | See ioengine=net for more. If the ioengine is file based, you |
| 334 | can specify a number of files by separating the names with a |
| 335 | ':' colon. So if you wanted a job to open /dev/sda and /dev/sdb |
| 336 | as the two working files, you would use |
| 337 | filename=/dev/sda:/dev/sdb. On Windows, disk devices are |
| 338 | accessed as \\.\PhysicalDrive0 for the first device, |
| 339 | \\.\PhysicalDrive1 for the second etc. Note: Windows and |
| 340 | FreeBSD prevent write access to areas of the disk containing |
| 341 | in-use data (e.g. filesystems). |
| 342 | If the wanted filename does need to include a colon, then |
| 343 | escape that with a '\' character. For instance, if the filename |
| 344 | is "/dev/dsk/foo@3,0:c", then you would use |
| 345 | filename="/dev/dsk/foo@3,0\:c". '-' is a reserved name, meaning |
| 346 | stdin or stdout. Which of the two depends on the read/write |
| 347 | direction set. |
| 348 | |
| 349 | filename_format=str |
| 350 | If sharing multiple files between jobs, it is usually necessary |
| 351 | to have fio generate the exact names that you want. By default, |
| 352 | fio will name a file based on the default file format |
| 353 | specification of jobname.jobnumber.filenumber. With this |
| 354 | option, that can be customized. Fio will recognize and replace |
| 355 | the following keywords in this string: |
| 356 | |
| 357 | $jobname |
| 358 | The name of the worker thread or process. |
| 359 | |
| 360 | $jobnum |
| 361 | The incremental number of the worker thread or |
| 362 | process. |
| 363 | |
| 364 | $filenum |
| 365 | The incremental number of the file for that worker |
| 366 | thread or process. |
| 367 | |
| 368 | To have dependent jobs share a set of files, this option can |
| 369 | be set to have fio generate filenames that are shared between |
| 370 | the two. For instance, if testfiles.$filenum is specified, |
| 371 | file number 4 for any job will be named testfiles.4. The |
| 372 | default of $jobname.$jobnum.$filenum will be used if |
| 373 | no other format specifier is given. |
| 374 | |
| 375 | opendir=str Tell fio to recursively add any file it can find in this |
| 376 | directory and down the file system tree. |
| 377 | |
| 378 | lockfile=str Fio defaults to not locking any files before it does |
| 379 | IO to them. If a file or file descriptor is shared, fio |
| 380 | can serialize IO to that file to make the end result |
| 381 | consistent. This is usual for emulating real workloads that |
| 382 | share files. The lock modes are: |
| 383 | |
| 384 | none No locking. The default. |
| 385 | exclusive Only one thread/process may do IO, |
| 386 | excluding all others. |
| 387 | readwrite Read-write locking on the file. Many |
| 388 | readers may access the file at the |
| 389 | same time, but writes get exclusive |
| 390 | access. |
| 391 | |
| 392 | readwrite=str |
| 393 | rw=str Type of io pattern. Accepted values are: |
| 394 | |
| 395 | read Sequential reads |
| 396 | write Sequential writes |
| 397 | randwrite Random writes |
| 398 | randread Random reads |
| 399 | rw,readwrite Sequential mixed reads and writes |
| 400 | randrw Random mixed reads and writes |
| 401 | trimwrite Mixed trims and writes. Blocks will be |
| 402 | trimmed first, then written to. |
| 403 | |
| 404 | For the mixed io types, the default is to split them 50/50. |
| 405 | For certain types of io the result may still be skewed a bit, |
| 406 | since the speed may be different. It is possible to specify |
| 407 | a number of IO's to do before getting a new offset, this is |
| 408 | done by appending a ':<nr>' to the end of the string given. |
| 409 | For a random read, it would look like 'rw=randread:8' for |
| 410 | passing in an offset modifier with a value of 8. If the |
| 411 | suffix is used with a sequential IO pattern, then the value |
| 412 | specified will be added to the generated offset for each IO. |
| 413 | For instance, using rw=write:4k will skip 4k for every |
| 414 | write. It turns sequential IO into sequential IO with holes. |
| 415 | See the 'rw_sequencer' option. |
| 416 | |
| 417 | rw_sequencer=str If an offset modifier is given by appending a number to |
| 418 | the rw=<str> line, then this option controls how that |
| 419 | number modifies the IO offset being generated. Accepted |
| 420 | values are: |
| 421 | |
| 422 | sequential Generate sequential offset |
| 423 | identical Generate the same offset |
| 424 | |
| 425 | 'sequential' is only useful for random IO, where fio would |
| 426 | normally generate a new random offset for every IO. If you |
| 427 | append eg 8 to randread, you would get a new random offset for |
| 428 | every 8 IO's. The result would be a seek for only every 8 |
| 429 | IO's, instead of for every IO. Use rw=randread:8 to specify |
| 430 | that. As sequential IO is already sequential, setting |
| 431 | 'sequential' for that would not result in any differences. |
| 432 | 'identical' behaves in a similar fashion, except it sends |
| 433 | the same offset 8 number of times before generating a new |
| 434 | offset. |
| 435 | |
| 436 | kb_base=int The base unit for a kilobyte. The defacto base is 2^10, 1024. |
| 437 | Storage manufacturers like to use 10^3 or 1000 as a base |
| 438 | ten unit instead, for obvious reasons. Allow values are |
| 439 | 1024 or 1000, with 1024 being the default. |
| 440 | |
| 441 | unified_rw_reporting=bool Fio normally reports statistics on a per |
| 442 | data direction basis, meaning that read, write, and trim are |
| 443 | accounted and reported separately. If this option is set, |
| 444 | the fio will sum the results and report them as "mixed" |
| 445 | instead. |
| 446 | |
| 447 | randrepeat=bool For random IO workloads, seed the generator in a predictable |
| 448 | way so that results are repeatable across repetitions. |
| 449 | Defaults to true. |
| 450 | |
| 451 | randseed=int Seed the random number generators based on this seed value, to |
| 452 | be able to control what sequence of output is being generated. |
| 453 | If not set, the random sequence depends on the randrepeat |
| 454 | setting. |
| 455 | |
| 456 | fallocate=str Whether pre-allocation is performed when laying down files. |
| 457 | Accepted values are: |
| 458 | |
| 459 | none Do not pre-allocate space |
| 460 | posix Pre-allocate via posix_fallocate() |
| 461 | keep Pre-allocate via fallocate() with |
| 462 | FALLOC_FL_KEEP_SIZE set |
| 463 | 0 Backward-compatible alias for 'none' |
| 464 | 1 Backward-compatible alias for 'posix' |
| 465 | |
| 466 | May not be available on all supported platforms. 'keep' is only |
| 467 | available on Linux.If using ZFS on Solaris this must be set to |
| 468 | 'none' because ZFS doesn't support it. Default: 'posix'. |
| 469 | |
| 470 | fadvise_hint=bool By default, fio will use fadvise() to advise the kernel |
| 471 | on what IO patterns it is likely to issue. Sometimes you |
| 472 | want to test specific IO patterns without telling the |
| 473 | kernel about it, in which case you can disable this option. |
| 474 | If set, fio will use POSIX_FADV_SEQUENTIAL for sequential |
| 475 | IO and POSIX_FADV_RANDOM for random IO. |
| 476 | |
| 477 | fadvise_stream=int Notify the kernel what write stream ID to place these |
| 478 | writes under. Only supported on Linux. Note, this option |
| 479 | may change going forward. |
| 480 | |
| 481 | size=int The total size of file io for this job. Fio will run until |
| 482 | this many bytes has been transferred, unless runtime is |
| 483 | limited by other options (such as 'runtime', for instance, |
| 484 | or increased/decreased by 'io_size'). Unless specific nrfiles |
| 485 | and filesize options are given, fio will divide this size |
| 486 | between the available files specified by the job. If not set, |
| 487 | fio will use the full size of the given files or devices. |
| 488 | If the files do not exist, size must be given. It is also |
| 489 | possible to give size as a percentage between 1 and 100. If |
| 490 | size=20% is given, fio will use 20% of the full size of the |
| 491 | given files or devices. |
| 492 | |
| 493 | io_size=int |
| 494 | io_limit=int Normally fio operates within the region set by 'size', which |
| 495 | means that the 'size' option sets both the region and size of |
| 496 | IO to be performed. Sometimes that is not what you want. With |
| 497 | this option, it is possible to define just the amount of IO |
| 498 | that fio should do. For instance, if 'size' is set to 20G and |
| 499 | 'io_size' is set to 5G, fio will perform IO within the first |
| 500 | 20G but exit when 5G have been done. The opposite is also |
| 501 | possible - if 'size' is set to 20G, and 'io_size' is set to |
| 502 | 40G, then fio will do 40G of IO within the 0..20G region. |
| 503 | |
| 504 | filesize=int Individual file sizes. May be a range, in which case fio |
| 505 | will select sizes for files at random within the given range |
| 506 | and limited to 'size' in total (if that is given). If not |
| 507 | given, each created file is the same size. |
| 508 | |
| 509 | file_append=bool Perform IO after the end of the file. Normally fio will |
| 510 | operate within the size of a file. If this option is set, then |
| 511 | fio will append to the file instead. This has identical |
| 512 | behavior to setting offset to the size of a file. This option |
| 513 | is ignored on non-regular files. |
| 514 | |
| 515 | fill_device=bool |
| 516 | fill_fs=bool Sets size to something really large and waits for ENOSPC (no |
| 517 | space left on device) as the terminating condition. Only makes |
| 518 | sense with sequential write. For a read workload, the mount |
| 519 | point will be filled first then IO started on the result. This |
| 520 | option doesn't make sense if operating on a raw device node, |
| 521 | since the size of that is already known by the file system. |
| 522 | Additionally, writing beyond end-of-device will not return |
| 523 | ENOSPC there. |
| 524 | |
| 525 | blocksize=int |
| 526 | bs=int The block size used for the io units. Defaults to 4k. Values |
| 527 | can be given for both read and writes. If a single int is |
| 528 | given, it will apply to both. If a second int is specified |
| 529 | after a comma, it will apply to writes only. In other words, |
| 530 | the format is either bs=read_and_write or bs=read,write,trim. |
| 531 | bs=4k,8k will thus use 4k blocks for reads, 8k blocks for |
| 532 | writes, and 8k for trims. You can terminate the list with |
| 533 | a trailing comma. bs=4k,8k, would use the default value for |
| 534 | trims.. If you only wish to set the write size, you |
| 535 | can do so by passing an empty read size - bs=,8k will set |
| 536 | 8k for writes and leave the read default value. |
| 537 | |
| 538 | blockalign=int |
| 539 | ba=int At what boundary to align random IO offsets. Defaults to |
| 540 | the same as 'blocksize' the minimum blocksize given. |
| 541 | Minimum alignment is typically 512b for using direct IO, |
| 542 | though it usually depends on the hardware block size. This |
| 543 | option is mutually exclusive with using a random map for |
| 544 | files, so it will turn off that option. |
| 545 | |
| 546 | blocksize_range=irange |
| 547 | bsrange=irange Instead of giving a single block size, specify a range |
| 548 | and fio will mix the issued io block sizes. The issued |
| 549 | io unit will always be a multiple of the minimum value |
| 550 | given (also see bs_unaligned). Applies to both reads and |
| 551 | writes, however a second range can be given after a comma. |
| 552 | See bs=. |
| 553 | |
| 554 | bssplit=str Sometimes you want even finer grained control of the |
| 555 | block sizes issued, not just an even split between them. |
| 556 | This option allows you to weight various block sizes, |
| 557 | so that you are able to define a specific amount of |
| 558 | block sizes issued. The format for this option is: |
| 559 | |
| 560 | bssplit=blocksize/percentage:blocksize/percentage |
| 561 | |
| 562 | for as many block sizes as needed. So if you want to define |
| 563 | a workload that has 50% 64k blocks, 10% 4k blocks, and |
| 564 | 40% 32k blocks, you would write: |
| 565 | |
| 566 | bssplit=4k/10:64k/50:32k/40 |
| 567 | |
| 568 | Ordering does not matter. If the percentage is left blank, |
| 569 | fio will fill in the remaining values evenly. So a bssplit |
| 570 | option like this one: |
| 571 | |
| 572 | bssplit=4k/50:1k/:32k/ |
| 573 | |
| 574 | would have 50% 4k ios, and 25% 1k and 32k ios. The percentages |
| 575 | always add up to 100, if bssplit is given a range that adds |
| 576 | up to more, it will error out. |
| 577 | |
| 578 | bssplit also supports giving separate splits to reads and |
| 579 | writes. The format is identical to what bs= accepts. You |
| 580 | have to separate the read and write parts with a comma. So |
| 581 | if you want a workload that has 50% 2k reads and 50% 4k reads, |
| 582 | while having 90% 4k writes and 10% 8k writes, you would |
| 583 | specify: |
| 584 | |
| 585 | bssplit=2k/50:4k/50,4k/90:8k/10 |
| 586 | |
| 587 | blocksize_unaligned |
| 588 | bs_unaligned If this option is given, any byte size value within bsrange |
| 589 | may be used as a block range. This typically wont work with |
| 590 | direct IO, as that normally requires sector alignment. |
| 591 | |
| 592 | bs_is_seq_rand If this option is set, fio will use the normal read,write |
| 593 | blocksize settings as sequential,random instead. Any random |
| 594 | read or write will use the WRITE blocksize settings, and any |
| 595 | sequential read or write will use the READ blocksize setting. |
| 596 | |
| 597 | zero_buffers If this option is given, fio will init the IO buffers to |
| 598 | all zeroes. The default is to fill them with random data. |
| 599 | |
| 600 | refill_buffers If this option is given, fio will refill the IO buffers |
| 601 | on every submit. The default is to only fill it at init |
| 602 | time and reuse that data. Only makes sense if zero_buffers |
| 603 | isn't specified, naturally. If data verification is enabled, |
| 604 | refill_buffers is also automatically enabled. |
| 605 | |
| 606 | scramble_buffers=bool If refill_buffers is too costly and the target is |
| 607 | using data deduplication, then setting this option will |
| 608 | slightly modify the IO buffer contents to defeat normal |
| 609 | de-dupe attempts. This is not enough to defeat more clever |
| 610 | block compression attempts, but it will stop naive dedupe of |
| 611 | blocks. Default: true. |
| 612 | |
| 613 | buffer_compress_percentage=int If this is set, then fio will attempt to |
| 614 | provide IO buffer content (on WRITEs) that compress to |
| 615 | the specified level. Fio does this by providing a mix of |
| 616 | random data and a fixed pattern. The fixed pattern is either |
| 617 | zeroes, or the pattern specified by buffer_pattern. If the |
| 618 | pattern option is used, it might skew the compression ratio |
| 619 | slightly. Note that this is per block size unit, for file/disk |
| 620 | wide compression level that matches this setting, you'll also |
| 621 | want to set refill_buffers. |
| 622 | |
| 623 | buffer_compress_chunk=int See buffer_compress_percentage. This |
| 624 | setting allows fio to manage how big the ranges of random |
| 625 | data and zeroed data is. Without this set, fio will |
| 626 | provide buffer_compress_percentage of blocksize random |
| 627 | data, followed by the remaining zeroed. With this set |
| 628 | to some chunk size smaller than the block size, fio can |
| 629 | alternate random and zeroed data throughout the IO |
| 630 | buffer. |
| 631 | |
| 632 | buffer_pattern=str If set, fio will fill the io buffers with this |
| 633 | pattern. If not set, the contents of io buffers is defined by |
| 634 | the other options related to buffer contents. The setting can |
| 635 | be any pattern of bytes, and can be prefixed with 0x for hex |
| 636 | values. It may also be a string, where the string must then |
| 637 | be wrapped with "", e.g.: |
| 638 | |
| 639 | buffer_pattern="abcd" |
| 640 | or |
| 641 | buffer_pattern=-12 |
| 642 | or |
| 643 | buffer_pattern=0xdeadface |
| 644 | |
| 645 | Also you can combine everything together in any order: |
| 646 | buffer_pattern=0xdeadface"abcd"-12 |
| 647 | |
| 648 | dedupe_percentage=int If set, fio will generate this percentage of |
| 649 | identical buffers when writing. These buffers will be |
| 650 | naturally dedupable. The contents of the buffers depend on |
| 651 | what other buffer compression settings have been set. It's |
| 652 | possible to have the individual buffers either fully |
| 653 | compressible, or not at all. This option only controls the |
| 654 | distribution of unique buffers. |
| 655 | |
| 656 | nrfiles=int Number of files to use for this job. Defaults to 1. |
| 657 | |
| 658 | openfiles=int Number of files to keep open at the same time. Defaults to |
| 659 | the same as nrfiles, can be set smaller to limit the number |
| 660 | simultaneous opens. |
| 661 | |
| 662 | file_service_type=str Defines how fio decides which file from a job to |
| 663 | service next. The following types are defined: |
| 664 | |
| 665 | random Just choose a file at random. |
| 666 | |
| 667 | roundrobin Round robin over open files. This |
| 668 | is the default. |
| 669 | |
| 670 | sequential Finish one file before moving on to |
| 671 | the next. Multiple files can still be |
| 672 | open depending on 'openfiles'. |
| 673 | |
| 674 | The string can have a number appended, indicating how |
| 675 | often to switch to a new file. So if option random:4 is |
| 676 | given, fio will switch to a new random file after 4 ios |
| 677 | have been issued. |
| 678 | |
| 679 | ioengine=str Defines how the job issues io to the file. The following |
| 680 | types are defined: |
| 681 | |
| 682 | sync Basic read(2) or write(2) io. lseek(2) is |
| 683 | used to position the io location. |
| 684 | |
| 685 | psync Basic pread(2) or pwrite(2) io. |
| 686 | |
| 687 | vsync Basic readv(2) or writev(2) IO. |
| 688 | |
| 689 | psyncv Basic preadv(2) or pwritev(2) IO. |
| 690 | |
| 691 | libaio Linux native asynchronous io. Note that Linux |
| 692 | may only support queued behaviour with |
| 693 | non-buffered IO (set direct=1 or buffered=0). |
| 694 | This engine defines engine specific options. |
| 695 | |
| 696 | posixaio glibc posix asynchronous io. |
| 697 | |
| 698 | solarisaio Solaris native asynchronous io. |
| 699 | |
| 700 | windowsaio Windows native asynchronous io. |
| 701 | |
| 702 | mmap File is memory mapped and data copied |
| 703 | to/from using memcpy(3). |
| 704 | |
| 705 | splice splice(2) is used to transfer the data and |
| 706 | vmsplice(2) to transfer data from user |
| 707 | space to the kernel. |
| 708 | |
| 709 | syslet-rw Use the syslet system calls to make |
| 710 | regular read/write async. |
| 711 | |
| 712 | sg SCSI generic sg v3 io. May either be |
| 713 | synchronous using the SG_IO ioctl, or if |
| 714 | the target is an sg character device |
| 715 | we use read(2) and write(2) for asynchronous |
| 716 | io. |
| 717 | |
| 718 | null Doesn't transfer any data, just pretends |
| 719 | to. This is mainly used to exercise fio |
| 720 | itself and for debugging/testing purposes. |
| 721 | |
| 722 | net Transfer over the network to given host:port. |
| 723 | Depending on the protocol used, the hostname, |
| 724 | port, listen and filename options are used to |
| 725 | specify what sort of connection to make, while |
| 726 | the protocol option determines which protocol |
| 727 | will be used. |
| 728 | This engine defines engine specific options. |
| 729 | |
| 730 | netsplice Like net, but uses splice/vmsplice to |
| 731 | map data and send/receive. |
| 732 | This engine defines engine specific options. |
| 733 | |
| 734 | cpuio Doesn't transfer any data, but burns CPU |
| 735 | cycles according to the cpuload= and |
| 736 | cpucycle= options. Setting cpuload=85 |
| 737 | will cause that job to do nothing but burn |
| 738 | 85% of the CPU. In case of SMP machines, |
| 739 | use numjobs=<no_of_cpu> to get desired CPU |
| 740 | usage, as the cpuload only loads a single |
| 741 | CPU at the desired rate. |
| 742 | |
| 743 | guasi The GUASI IO engine is the Generic Userspace |
| 744 | Asyncronous Syscall Interface approach |
| 745 | to async IO. See |
| 746 | |
| 747 | http://www.xmailserver.org/guasi-lib.html |
| 748 | |
| 749 | for more info on GUASI. |
| 750 | |
| 751 | rdma The RDMA I/O engine supports both RDMA |
| 752 | memory semantics (RDMA_WRITE/RDMA_READ) and |
| 753 | channel semantics (Send/Recv) for the |
| 754 | InfiniBand, RoCE and iWARP protocols. |
| 755 | |
| 756 | falloc IO engine that does regular fallocate to |
| 757 | simulate data transfer as fio ioengine. |
| 758 | DDIR_READ does fallocate(,mode = keep_size,) |
| 759 | DDIR_WRITE does fallocate(,mode = 0) |
| 760 | DDIR_TRIM does fallocate(,mode = punch_hole) |
| 761 | |
| 762 | e4defrag IO engine that does regular EXT4_IOC_MOVE_EXT |
| 763 | ioctls to simulate defragment activity in |
| 764 | request to DDIR_WRITE event |
| 765 | |
| 766 | rbd IO engine supporting direct access to Ceph |
| 767 | Rados Block Devices (RBD) via librbd without |
| 768 | the need to use the kernel rbd driver. This |
| 769 | ioengine defines engine specific options. |
| 770 | |
| 771 | gfapi Using Glusterfs libgfapi sync interface to |
| 772 | direct access to Glusterfs volumes without |
| 773 | options. |
| 774 | |
| 775 | gfapi_async Using Glusterfs libgfapi async interface |
| 776 | to direct access to Glusterfs volumes without |
| 777 | having to go through FUSE. This ioengine |
| 778 | defines engine specific options. |
| 779 | |
| 780 | libhdfs Read and write through Hadoop (HDFS). |
| 781 | This engine interprets offsets a little |
| 782 | differently. In HDFS, files once created |
| 783 | cannot be modified. So random writes are not |
| 784 | possible. To imitate this, libhdfs engine |
| 785 | creates bunch of small files, and engine will |
| 786 | pick a file out of those files based on the |
| 787 | offset enerated by fio backend. Each jobs uses |
| 788 | it's own connection to HDFS. |
| 789 | |
| 790 | mtd Read, write and erase an MTD character device |
| 791 | (e.g., /dev/mtd0). Discards are treated as |
| 792 | erases. Depending on the underlying device |
| 793 | type, the I/O may have to go in a certain |
| 794 | pattern, e.g., on NAND, writing sequentially |
| 795 | to erase blocks and discarding before |
| 796 | overwriting. The writetrim mode works well |
| 797 | for this constraint. |
| 798 | |
| 799 | external Prefix to specify loading an external |
| 800 | IO engine object file. Append the engine |
| 801 | filename, eg ioengine=external:/tmp/foo.o |
| 802 | to load ioengine foo.o in /tmp. |
| 803 | |
| 804 | iodepth=int This defines how many io units to keep in flight against |
| 805 | the file. The default is 1 for each file defined in this |
| 806 | job, can be overridden with a larger value for higher |
| 807 | concurrency. Note that increasing iodepth beyond 1 will not |
| 808 | affect synchronous ioengines (except for small degress when |
| 809 | verify_async is in use). Even async engines may impose OS |
| 810 | restrictions causing the desired depth not to be achieved. |
| 811 | This may happen on Linux when using libaio and not setting |
| 812 | direct=1, since buffered IO is not async on that OS. Keep an |
| 813 | eye on the IO depth distribution in the fio output to verify |
| 814 | that the achieved depth is as expected. Default: 1. |
| 815 | |
| 816 | iodepth_batch_submit=int |
| 817 | iodepth_batch=int This defines how many pieces of IO to submit at once. |
| 818 | It defaults to 1 which means that we submit each IO |
| 819 | as soon as it is available, but can be raised to submit |
| 820 | bigger batches of IO at the time. If it is set to 0 the iodepth |
| 821 | value will be used. |
| 822 | |
| 823 | iodepth_batch_complete_min=int |
| 824 | iodepth_batch_complete=int This defines how many pieces of IO to retrieve |
| 825 | at once. It defaults to 1 which means that we'll ask |
| 826 | for a minimum of 1 IO in the retrieval process from |
| 827 | the kernel. The IO retrieval will go on until we |
| 828 | hit the limit set by iodepth_low. If this variable is |
| 829 | set to 0, then fio will always check for completed |
| 830 | events before queuing more IO. This helps reduce |
| 831 | IO latency, at the cost of more retrieval system calls. |
| 832 | |
| 833 | iodepth_batch_complete_max=int This defines maximum pieces of IO to |
| 834 | retrieve at once. This variable should be used along with |
| 835 | iodepth_batch_complete_min=int variable, specifying the range |
| 836 | of min and max amount of IO which should be retrieved. By default |
| 837 | it is equal to iodepth_batch_complete_min value. |
| 838 | |
| 839 | Example #1: |
| 840 | |
| 841 | iodepth_batch_complete_min=1 |
| 842 | iodepth_batch_complete_max=<iodepth> |
| 843 | |
| 844 | which means that we will retrieve at leat 1 IO and up to the |
| 845 | whole submitted queue depth. If none of IO has been completed |
| 846 | yet, we will wait. |
| 847 | |
| 848 | Example #2: |
| 849 | |
| 850 | iodepth_batch_complete_min=0 |
| 851 | iodepth_batch_complete_max=<iodepth> |
| 852 | |
| 853 | which means that we can retrieve up to the whole submitted |
| 854 | queue depth, but if none of IO has been completed yet, we will |
| 855 | NOT wait and immediately exit the system call. In this example |
| 856 | we simply do polling. |
| 857 | |
| 858 | iodepth_low=int The low water mark indicating when to start filling |
| 859 | the queue again. Defaults to the same as iodepth, meaning |
| 860 | that fio will attempt to keep the queue full at all times. |
| 861 | If iodepth is set to eg 16 and iodepth_low is set to 4, then |
| 862 | after fio has filled the queue of 16 requests, it will let |
| 863 | the depth drain down to 4 before starting to fill it again. |
| 864 | |
| 865 | io_submit_mode=str This option controls how fio submits the IO to |
| 866 | the IO engine. The default is 'inline', which means that the |
| 867 | fio job threads submit and reap IO directly. If set to |
| 868 | 'offload', the job threads will offload IO submission to a |
| 869 | dedicated pool of IO threads. This requires some coordination |
| 870 | and thus has a bit of extra overhead, especially for lower |
| 871 | queue depth IO where it can increase latencies. The benefit |
| 872 | is that fio can manage submission rates independently of |
| 873 | the device completion rates. This avoids skewed latency |
| 874 | reporting if IO gets back up on the device side (the |
| 875 | coordinated omission problem). |
| 876 | |
| 877 | direct=bool If value is true, use non-buffered io. This is usually |
| 878 | O_DIRECT. Note that ZFS on Solaris doesn't support direct io. |
| 879 | On Windows the synchronous ioengines don't support direct io. |
| 880 | |
| 881 | atomic=bool If value is true, attempt to use atomic direct IO. Atomic |
| 882 | writes are guaranteed to be stable once acknowledged by |
| 883 | the operating system. Only Linux supports O_ATOMIC right |
| 884 | now. |
| 885 | |
| 886 | buffered=bool If value is true, use buffered io. This is the opposite |
| 887 | of the 'direct' option. Defaults to true. |
| 888 | |
| 889 | offset=int Start io at the given offset in the file. The data before |
| 890 | the given offset will not be touched. This effectively |
| 891 | caps the file size at real_size - offset. |
| 892 | |
| 893 | offset_increment=int If this is provided, then the real offset becomes |
| 894 | offset + offset_increment * thread_number, where the thread |
| 895 | number is a counter that starts at 0 and is incremented for |
| 896 | each sub-job (i.e. when numjobs option is specified). This |
| 897 | option is useful if there are several jobs which are intended |
| 898 | to operate on a file in parallel disjoint segments, with |
| 899 | even spacing between the starting points. |
| 900 | |
| 901 | number_ios=int Fio will normally perform IOs until it has exhausted the size |
| 902 | of the region set by size=, or if it exhaust the allocated |
| 903 | time (or hits an error condition). With this setting, the |
| 904 | range/size can be set independently of the number of IOs to |
| 905 | perform. When fio reaches this number, it will exit normally |
| 906 | and report status. Note that this does not extend the amount |
| 907 | of IO that will be done, it will only stop fio if this |
| 908 | condition is met before other end-of-job criteria. |
| 909 | |
| 910 | fsync=int If writing to a file, issue a sync of the dirty data |
| 911 | for every number of blocks given. For example, if you give |
| 912 | 32 as a parameter, fio will sync the file for every 32 |
| 913 | writes issued. If fio is using non-buffered io, we may |
| 914 | not sync the file. The exception is the sg io engine, which |
| 915 | synchronizes the disk cache anyway. |
| 916 | |
| 917 | fdatasync=int Like fsync= but uses fdatasync() to only sync data and not |
| 918 | metadata blocks. |
| 919 | In FreeBSD and Windows there is no fdatasync(), this falls back |
| 920 | to using fsync() |
| 921 | |
| 922 | sync_file_range=str:val Use sync_file_range() for every 'val' number of |
| 923 | write operations. Fio will track range of writes that |
| 924 | have happened since the last sync_file_range() call. 'str' |
| 925 | can currently be one or more of: |
| 926 | |
| 927 | wait_before SYNC_FILE_RANGE_WAIT_BEFORE |
| 928 | write SYNC_FILE_RANGE_WRITE |
| 929 | wait_after SYNC_FILE_RANGE_WAIT_AFTER |
| 930 | |
| 931 | So if you do sync_file_range=wait_before,write:8, fio would |
| 932 | use SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE for |
| 933 | every 8 writes. Also see the sync_file_range(2) man page. |
| 934 | This option is Linux specific. |
| 935 | |
| 936 | overwrite=bool If true, writes to a file will always overwrite existing |
| 937 | data. If the file doesn't already exist, it will be |
| 938 | created before the write phase begins. If the file exists |
| 939 | and is large enough for the specified write phase, nothing |
| 940 | will be done. |
| 941 | |
| 942 | end_fsync=bool If true, fsync file contents when a write stage has completed. |
| 943 | |
| 944 | fsync_on_close=bool If true, fio will fsync() a dirty file on close. |
| 945 | This differs from end_fsync in that it will happen on every |
| 946 | file close, not just at the end of the job. |
| 947 | |
| 948 | rwmixread=int How large a percentage of the mix should be reads. |
| 949 | |
| 950 | rwmixwrite=int How large a percentage of the mix should be writes. If both |
| 951 | rwmixread and rwmixwrite is given and the values do not add |
| 952 | up to 100%, the latter of the two will be used to override |
| 953 | the first. This may interfere with a given rate setting, |
| 954 | if fio is asked to limit reads or writes to a certain rate. |
| 955 | If that is the case, then the distribution may be skewed. |
| 956 | |
| 957 | random_distribution=str:float By default, fio will use a completely uniform |
| 958 | random distribution when asked to perform random IO. Sometimes |
| 959 | it is useful to skew the distribution in specific ways, |
| 960 | ensuring that some parts of the data is more hot than others. |
| 961 | fio includes the following distribution models: |
| 962 | |
| 963 | random Uniform random distribution |
| 964 | zipf Zipf distribution |
| 965 | pareto Pareto distribution |
| 966 | |
| 967 | When using a zipf or pareto distribution, an input value |
| 968 | is also needed to define the access pattern. For zipf, this |
| 969 | is the zipf theta. For pareto, it's the pareto power. Fio |
| 970 | includes a test program, genzipf, that can be used visualize |
| 971 | what the given input values will yield in terms of hit rates. |
| 972 | If you wanted to use zipf with a theta of 1.2, you would use |
| 973 | random_distribution=zipf:1.2 as the option. If a non-uniform |
| 974 | model is used, fio will disable use of the random map. |
| 975 | |
| 976 | percentage_random=int For a random workload, set how big a percentage should |
| 977 | be random. This defaults to 100%, in which case the workload |
| 978 | is fully random. It can be set from anywhere from 0 to 100. |
| 979 | Setting it to 0 would make the workload fully sequential. Any |
| 980 | setting in between will result in a random mix of sequential |
| 981 | and random IO, at the given percentages. It is possible to |
| 982 | set different values for reads, writes, and trim. To do so, |
| 983 | simply use a comma separated list. See blocksize. |
| 984 | |
| 985 | norandommap Normally fio will cover every block of the file when doing |
| 986 | random IO. If this option is given, fio will just get a |
| 987 | new random offset without looking at past io history. This |
| 988 | means that some blocks may not be read or written, and that |
| 989 | some blocks may be read/written more than once. If this option |
| 990 | is used with verify= and multiple blocksizes (via bsrange=), |
| 991 | only intact blocks are verified, i.e., partially-overwritten |
| 992 | blocks are ignored. |
| 993 | |
| 994 | softrandommap=bool See norandommap. If fio runs with the random block map |
| 995 | enabled and it fails to allocate the map, if this option is |
| 996 | set it will continue without a random block map. As coverage |
| 997 | will not be as complete as with random maps, this option is |
| 998 | disabled by default. |
| 999 | |
| 1000 | random_generator=str Fio supports the following engines for generating |
| 1001 | IO offsets for random IO: |
| 1002 | |
| 1003 | tausworthe Strong 2^88 cycle random number generator |
| 1004 | lfsr Linear feedback shift register generator |
| 1005 | tausworthe64 Strong 64-bit 2^258 cycle random number |
| 1006 | generator |
| 1007 | |
| 1008 | Tausworthe is a strong random number generator, but it |
| 1009 | requires tracking on the side if we want to ensure that |
| 1010 | blocks are only read or written once. LFSR guarantees |
| 1011 | that we never generate the same offset twice, and it's |
| 1012 | also less computationally expensive. It's not a true |
| 1013 | random generator, however, though for IO purposes it's |
| 1014 | typically good enough. LFSR only works with single |
| 1015 | block sizes, not with workloads that use multiple block |
| 1016 | sizes. If used with such a workload, fio may read or write |
| 1017 | some blocks multiple times. The default value is tausworthe, |
| 1018 | unless the required space exceeds 2^32 blocks. If it does, |
| 1019 | then tausworthe64 is selected automatically. |
| 1020 | |
| 1021 | nice=int Run the job with the given nice value. See man nice(2). |
| 1022 | |
| 1023 | prio=int Set the io priority value of this job. Linux limits us to |
| 1024 | a positive value between 0 and 7, with 0 being the highest. |
| 1025 | See man ionice(1). |
| 1026 | |
| 1027 | prioclass=int Set the io priority class. See man ionice(1). |
| 1028 | |
| 1029 | thinktime=int Stall the job x microseconds after an io has completed before |
| 1030 | issuing the next. May be used to simulate processing being |
| 1031 | done by an application. See thinktime_blocks and |
| 1032 | thinktime_spin. |
| 1033 | |
| 1034 | thinktime_spin=int |
| 1035 | Only valid if thinktime is set - pretend to spend CPU time |
| 1036 | doing something with the data received, before falling back |
| 1037 | to sleeping for the rest of the period specified by |
| 1038 | thinktime. |
| 1039 | |
| 1040 | thinktime_blocks=int |
| 1041 | Only valid if thinktime is set - control how many blocks |
| 1042 | to issue, before waiting 'thinktime' usecs. If not set, |
| 1043 | defaults to 1 which will make fio wait 'thinktime' usecs |
| 1044 | after every block. This effectively makes any queue depth |
| 1045 | setting redundant, since no more than 1 IO will be queued |
| 1046 | before we have to complete it and do our thinktime. In |
| 1047 | other words, this setting effectively caps the queue depth |
| 1048 | if the latter is larger. |
| 1049 | |
| 1050 | rate=int Cap the bandwidth used by this job. The number is in bytes/sec, |
| 1051 | the normal suffix rules apply. You can use rate=500k to limit |
| 1052 | reads and writes to 500k each, or you can specify read and |
| 1053 | writes separately. Using rate=1m,500k would limit reads to |
| 1054 | 1MB/sec and writes to 500KB/sec. Capping only reads or |
| 1055 | writes can be done with rate=,500k or rate=500k,. The former |
| 1056 | will only limit writes (to 500KB/sec), the latter will only |
| 1057 | limit reads. |
| 1058 | |
| 1059 | rate_min=int Tell fio to do whatever it can to maintain at least this |
| 1060 | bandwidth. Failing to meet this requirement, will cause |
| 1061 | the job to exit. The same format as rate is used for |
| 1062 | read vs write separation. |
| 1063 | |
| 1064 | rate_iops=int Cap the bandwidth to this number of IOPS. Basically the same |
| 1065 | as rate, just specified independently of bandwidth. If the |
| 1066 | job is given a block size range instead of a fixed value, |
| 1067 | the smallest block size is used as the metric. The same format |
| 1068 | as rate is used for read vs write separation. |
| 1069 | |
| 1070 | rate_iops_min=int If fio doesn't meet this rate of IO, it will cause |
| 1071 | the job to exit. The same format as rate is used for read vs |
| 1072 | write separation. |
| 1073 | |
| 1074 | rate_process=str This option controls how fio manages rated IO |
| 1075 | submissions. The default is 'linear', which submits IO in a |
| 1076 | linear fashion with fixed delays between IOs that gets |
| 1077 | adjusted based on IO completion rates. If this is set to |
| 1078 | 'poisson', fio will submit IO based on a more real world |
| 1079 | random request flow, known as the Poisson process |
| 1080 | (https://en.wikipedia.org/wiki/Poisson_process). The lambda |
| 1081 | will be 10^6 / IOPS for the given workload. |
| 1082 | |
| 1083 | latency_target=int If set, fio will attempt to find the max performance |
| 1084 | point that the given workload will run at while maintaining a |
| 1085 | latency below this target. The values is given in microseconds. |
| 1086 | See latency_window and latency_percentile |
| 1087 | |
| 1088 | latency_window=int Used with latency_target to specify the sample window |
| 1089 | that the job is run at varying queue depths to test the |
| 1090 | performance. The value is given in microseconds. |
| 1091 | |
| 1092 | latency_percentile=float The percentage of IOs that must fall within the |
| 1093 | criteria specified by latency_target and latency_window. If not |
| 1094 | set, this defaults to 100.0, meaning that all IOs must be equal |
| 1095 | or below to the value set by latency_target. |
| 1096 | |
| 1097 | max_latency=int If set, fio will exit the job if it exceeds this maximum |
| 1098 | latency. It will exit with an ETIME error. |
| 1099 | |
| 1100 | rate_cycle=int Average bandwidth for 'rate' and 'rate_min' over this number |
| 1101 | of milliseconds. |
| 1102 | |
| 1103 | cpumask=int Set the CPU affinity of this job. The parameter given is a |
| 1104 | bitmask of allowed CPU's the job may run on. So if you want |
| 1105 | the allowed CPUs to be 1 and 5, you would pass the decimal |
| 1106 | value of (1 << 1 | 1 << 5), or 34. See man |
| 1107 | sched_setaffinity(2). This may not work on all supported |
| 1108 | operating systems or kernel versions. This option doesn't |
| 1109 | work well for a higher CPU count than what you can store in |
| 1110 | an integer mask, so it can only control cpus 1-32. For |
| 1111 | boxes with larger CPU counts, use cpus_allowed. |
| 1112 | |
| 1113 | cpus_allowed=str Controls the same options as cpumask, but it allows a text |
| 1114 | setting of the permitted CPUs instead. So to use CPUs 1 and |
| 1115 | 5, you would specify cpus_allowed=1,5. This options also |
| 1116 | allows a range of CPUs. Say you wanted a binding to CPUs |
| 1117 | 1, 5, and 8-15, you would set cpus_allowed=1,5,8-15. |
| 1118 | |
| 1119 | cpus_allowed_policy=str Set the policy of how fio distributes the CPUs |
| 1120 | specified by cpus_allowed or cpumask. Two policies are |
| 1121 | supported: |
| 1122 | |
| 1123 | shared All jobs will share the CPU set specified. |
| 1124 | split Each job will get a unique CPU from the CPU set. |
| 1125 | |
| 1126 | 'shared' is the default behaviour, if the option isn't |
| 1127 | specified. If split is specified, then fio will will assign |
| 1128 | one cpu per job. If not enough CPUs are given for the jobs |
| 1129 | listed, then fio will roundrobin the CPUs in the set. |
| 1130 | |
| 1131 | numa_cpu_nodes=str Set this job running on spcified NUMA nodes' CPUs. The |
| 1132 | arguments allow comma delimited list of cpu numbers, |
| 1133 | A-B ranges, or 'all'. Note, to enable numa options support, |
| 1134 | fio must be built on a system with libnuma-dev(el) installed. |
| 1135 | |
| 1136 | numa_mem_policy=str Set this job's memory policy and corresponding NUMA |
| 1137 | nodes. Format of the argements: |
| 1138 | <mode>[:<nodelist>] |
| 1139 | `mode' is one of the following memory policy: |
| 1140 | default, prefer, bind, interleave, local |
| 1141 | For `default' and `local' memory policy, no node is |
| 1142 | needed to be specified. |
| 1143 | For `prefer', only one node is allowed. |
| 1144 | For `bind' and `interleave', it allow comma delimited |
| 1145 | list of numbers, A-B ranges, or 'all'. |
| 1146 | |
| 1147 | startdelay=time Start this job the specified number of seconds after fio |
| 1148 | has started. Only useful if the job file contains several |
| 1149 | jobs, and you want to delay starting some jobs to a certain |
| 1150 | time. |
| 1151 | |
| 1152 | runtime=time Tell fio to terminate processing after the specified number |
| 1153 | of seconds. It can be quite hard to determine for how long |
| 1154 | a specified job will run, so this parameter is handy to |
| 1155 | cap the total runtime to a given time. |
| 1156 | |
| 1157 | time_based If set, fio will run for the duration of the runtime |
| 1158 | specified even if the file(s) are completely read or |
| 1159 | written. It will simply loop over the same workload |
| 1160 | as many times as the runtime allows. |
| 1161 | |
| 1162 | ramp_time=time If set, fio will run the specified workload for this amount |
| 1163 | of time before logging any performance numbers. Useful for |
| 1164 | letting performance settle before logging results, thus |
| 1165 | minimizing the runtime required for stable results. Note |
| 1166 | that the ramp_time is considered lead in time for a job, |
| 1167 | thus it will increase the total runtime if a special timeout |
| 1168 | or runtime is specified. |
| 1169 | |
| 1170 | invalidate=bool Invalidate the buffer/page cache parts for this file prior |
| 1171 | to starting io. Defaults to true. |
| 1172 | |
| 1173 | sync=bool Use sync io for buffered writes. For the majority of the |
| 1174 | io engines, this means using O_SYNC. |
| 1175 | |
| 1176 | iomem=str |
| 1177 | mem=str Fio can use various types of memory as the io unit buffer. |
| 1178 | The allowed values are: |
| 1179 | |
| 1180 | malloc Use memory from malloc(3) as the buffers. |
| 1181 | |
| 1182 | shm Use shared memory as the buffers. Allocated |
| 1183 | through shmget(2). |
| 1184 | |
| 1185 | shmhuge Same as shm, but use huge pages as backing. |
| 1186 | |
| 1187 | mmap Use mmap to allocate buffers. May either be |
| 1188 | anonymous memory, or can be file backed if |
| 1189 | a filename is given after the option. The |
| 1190 | format is mem=mmap:/path/to/file. |
| 1191 | |
| 1192 | mmaphuge Use a memory mapped huge file as the buffer |
| 1193 | backing. Append filename after mmaphuge, ala |
| 1194 | mem=mmaphuge:/hugetlbfs/file |
| 1195 | |
| 1196 | mmapshared Same as mmap, but use a MMAP_SHARED |
| 1197 | mapping. |
| 1198 | |
| 1199 | The area allocated is a function of the maximum allowed |
| 1200 | bs size for the job, multiplied by the io depth given. Note |
| 1201 | that for shmhuge and mmaphuge to work, the system must have |
| 1202 | free huge pages allocated. This can normally be checked |
| 1203 | and set by reading/writing /proc/sys/vm/nr_hugepages on a |
| 1204 | Linux system. Fio assumes a huge page is 4MB in size. So |
| 1205 | to calculate the number of huge pages you need for a given |
| 1206 | job file, add up the io depth of all jobs (normally one unless |
| 1207 | iodepth= is used) and multiply by the maximum bs set. Then |
| 1208 | divide that number by the huge page size. You can see the |
| 1209 | size of the huge pages in /proc/meminfo. If no huge pages |
| 1210 | are allocated by having a non-zero number in nr_hugepages, |
| 1211 | using mmaphuge or shmhuge will fail. Also see hugepage-size. |
| 1212 | |
| 1213 | mmaphuge also needs to have hugetlbfs mounted and the file |
| 1214 | location should point there. So if it's mounted in /huge, |
| 1215 | you would use mem=mmaphuge:/huge/somefile. |
| 1216 | |
| 1217 | iomem_align=int This indiciates the memory alignment of the IO memory buffers. |
| 1218 | Note that the given alignment is applied to the first IO unit |
| 1219 | buffer, if using iodepth the alignment of the following buffers |
| 1220 | are given by the bs used. In other words, if using a bs that is |
| 1221 | a multiple of the page sized in the system, all buffers will |
| 1222 | be aligned to this value. If using a bs that is not page |
| 1223 | aligned, the alignment of subsequent IO memory buffers is the |
| 1224 | sum of the iomem_align and bs used. |
| 1225 | |
| 1226 | hugepage-size=int |
| 1227 | Defines the size of a huge page. Must at least be equal |
| 1228 | to the system setting, see /proc/meminfo. Defaults to 4MB. |
| 1229 | Should probably always be a multiple of megabytes, so using |
| 1230 | hugepage-size=Xm is the preferred way to set this to avoid |
| 1231 | setting a non-pow-2 bad value. |
| 1232 | |
| 1233 | exitall When one job finishes, terminate the rest. The default is |
| 1234 | to wait for each job to finish, sometimes that is not the |
| 1235 | desired action. |
| 1236 | |
| 1237 | exitall_on_error When one job finishes in error, terminate the rest. The |
| 1238 | default is to wait for each job to finish. |
| 1239 | |
| 1240 | bwavgtime=int Average the calculated bandwidth over the given time. Value |
| 1241 | is specified in milliseconds. |
| 1242 | |
| 1243 | iopsavgtime=int Average the calculated IOPS over the given time. Value |
| 1244 | is specified in milliseconds. |
| 1245 | |
| 1246 | create_serialize=bool If true, serialize the file creating for the jobs. |
| 1247 | This may be handy to avoid interleaving of data |
| 1248 | files, which may greatly depend on the filesystem |
| 1249 | used and even the number of processors in the system. |
| 1250 | |
| 1251 | create_fsync=bool fsync the data file after creation. This is the |
| 1252 | default. |
| 1253 | |
| 1254 | create_on_open=bool Don't pre-setup the files for IO, just create open() |
| 1255 | when it's time to do IO to that file. |
| 1256 | |
| 1257 | create_only=bool If true, fio will only run the setup phase of the job. |
| 1258 | If files need to be laid out or updated on disk, only |
| 1259 | that will be done. The actual job contents are not |
| 1260 | executed. |
| 1261 | |
| 1262 | allow_file_create=bool If true, fio is permitted to create files as part |
| 1263 | of its workload. This is the default behavior. If this |
| 1264 | option is false, then fio will error out if the files it |
| 1265 | needs to use don't already exist. Default: true. |
| 1266 | |
| 1267 | allow_mounted_write=bool If this isn't set, fio will abort jobs that |
| 1268 | are destructive (eg that write) to what appears to be a |
| 1269 | mounted device or partition. This should help catch creating |
| 1270 | inadvertently destructive tests, not realizing that the test |
| 1271 | will destroy data on the mounted file system. Default: false. |
| 1272 | |
| 1273 | pre_read=bool If this is given, files will be pre-read into memory before |
| 1274 | starting the given IO operation. This will also clear |
| 1275 | the 'invalidate' flag, since it is pointless to pre-read |
| 1276 | and then drop the cache. This will only work for IO engines |
| 1277 | that are seekable, since they allow you to read the same data |
| 1278 | multiple times. Thus it will not work on eg network or splice |
| 1279 | IO. |
| 1280 | |
| 1281 | unlink=bool Unlink the job files when done. Not the default, as repeated |
| 1282 | runs of that job would then waste time recreating the file |
| 1283 | set again and again. |
| 1284 | |
| 1285 | loops=int Run the specified number of iterations of this job. Used |
| 1286 | to repeat the same workload a given number of times. Defaults |
| 1287 | to 1. |
| 1288 | |
| 1289 | verify_only Do not perform specified workload---only verify data still |
| 1290 | matches previous invocation of this workload. This option |
| 1291 | allows one to check data multiple times at a later date |
| 1292 | without overwriting it. This option makes sense only for |
| 1293 | workloads that write data, and does not support workloads |
| 1294 | with the time_based option set. |
| 1295 | |
| 1296 | do_verify=bool Run the verify phase after a write phase. Only makes sense if |
| 1297 | verify is set. Defaults to 1. |
| 1298 | |
| 1299 | verify=str If writing to a file, fio can verify the file contents |
| 1300 | after each iteration of the job. Each verification method also implies |
| 1301 | verification of special header, which is written to the beginning of |
| 1302 | each block. This header also includes meta information, like offset |
| 1303 | of the block, block number, timestamp when block was written, etc. |
| 1304 | verify=str can be combined with verify_pattern=str option. |
| 1305 | The allowed values are: |
| 1306 | |
| 1307 | md5 Use an md5 sum of the data area and store |
| 1308 | it in the header of each block. |
| 1309 | |
| 1310 | crc64 Use an experimental crc64 sum of the data |
| 1311 | area and store it in the header of each |
| 1312 | block. |
| 1313 | |
| 1314 | crc32c Use a crc32c sum of the data area and store |
| 1315 | it in the header of each block. |
| 1316 | |
| 1317 | crc32c-intel Use hardware assisted crc32c calcuation |
| 1318 | provided on SSE4.2 enabled processors. Falls |
| 1319 | back to regular software crc32c, if not |
| 1320 | supported by the system. |
| 1321 | |
| 1322 | crc32 Use a crc32 sum of the data area and store |
| 1323 | it in the header of each block. |
| 1324 | |
| 1325 | crc16 Use a crc16 sum of the data area and store |
| 1326 | it in the header of each block. |
| 1327 | |
| 1328 | crc7 Use a crc7 sum of the data area and store |
| 1329 | it in the header of each block. |
| 1330 | |
| 1331 | xxhash Use xxhash as the checksum function. Generally |
| 1332 | the fastest software checksum that fio |
| 1333 | supports. |
| 1334 | |
| 1335 | sha512 Use sha512 as the checksum function. |
| 1336 | |
| 1337 | sha256 Use sha256 as the checksum function. |
| 1338 | |
| 1339 | sha1 Use optimized sha1 as the checksum function. |
| 1340 | |
| 1341 | meta This option is deprecated, since now meta information is |
| 1342 | included in generic verification header and meta verification |
| 1343 | happens by default. For detailed information see the description |
| 1344 | of the verify=str setting. This option is kept because of |
| 1345 | compatibility's sake with old configurations. Do not use it. |
| 1346 | |
| 1347 | pattern Verify a strict pattern. Normally fio includes |
| 1348 | a header with some basic information and |
| 1349 | checksumming, but if this option is set, only |
| 1350 | the specific pattern set with 'verify_pattern' |
| 1351 | is verified. |
| 1352 | |
| 1353 | null Only pretend to verify. Useful for testing |
| 1354 | internals with ioengine=null, not for much |
| 1355 | else. |
| 1356 | |
| 1357 | This option can be used for repeated burn-in tests of a |
| 1358 | system to make sure that the written data is also |
| 1359 | correctly read back. If the data direction given is |
| 1360 | a read or random read, fio will assume that it should |
| 1361 | verify a previously written file. If the data direction |
| 1362 | includes any form of write, the verify will be of the |
| 1363 | newly written data. |
| 1364 | |
| 1365 | verifysort=bool If set, fio will sort written verify blocks when it deems |
| 1366 | it faster to read them back in a sorted manner. This is |
| 1367 | often the case when overwriting an existing file, since |
| 1368 | the blocks are already laid out in the file system. You |
| 1369 | can ignore this option unless doing huge amounts of really |
| 1370 | fast IO where the red-black tree sorting CPU time becomes |
| 1371 | significant. |
| 1372 | |
| 1373 | verify_offset=int Swap the verification header with data somewhere else |
| 1374 | in the block before writing. Its swapped back before |
| 1375 | verifying. |
| 1376 | |
| 1377 | verify_interval=int Write the verification header at a finer granularity |
| 1378 | than the blocksize. It will be written for chunks the |
| 1379 | size of header_interval. blocksize should divide this |
| 1380 | evenly. |
| 1381 | |
| 1382 | verify_pattern=str If set, fio will fill the io buffers with this |
| 1383 | pattern. Fio defaults to filling with totally random |
| 1384 | bytes, but sometimes it's interesting to fill with a known |
| 1385 | pattern for io verification purposes. Depending on the |
| 1386 | width of the pattern, fio will fill 1/2/3/4 bytes of the |
| 1387 | buffer at the time(it can be either a decimal or a hex number). |
| 1388 | The verify_pattern if larger than a 32-bit quantity has to |
| 1389 | be a hex number that starts with either "0x" or "0X". Use |
| 1390 | with verify=str. Also, verify_pattern supports %o format, |
| 1391 | which means that for each block offset will be written and |
| 1392 | then verifyied back, e.g.: |
| 1393 | |
| 1394 | verify_pattern=%o |
| 1395 | |
| 1396 | Or use combination of everything: |
| 1397 | verify_pattern=0xff%o"abcd"-12 |
| 1398 | |
| 1399 | verify_fatal=bool Normally fio will keep checking the entire contents |
| 1400 | before quitting on a block verification failure. If this |
| 1401 | option is set, fio will exit the job on the first observed |
| 1402 | failure. |
| 1403 | |
| 1404 | verify_dump=bool If set, dump the contents of both the original data |
| 1405 | block and the data block we read off disk to files. This |
| 1406 | allows later analysis to inspect just what kind of data |
| 1407 | corruption occurred. Off by default. |
| 1408 | |
| 1409 | verify_async=int Fio will normally verify IO inline from the submitting |
| 1410 | thread. This option takes an integer describing how many |
| 1411 | async offload threads to create for IO verification instead, |
| 1412 | causing fio to offload the duty of verifying IO contents |
| 1413 | to one or more separate threads. If using this offload |
| 1414 | option, even sync IO engines can benefit from using an |
| 1415 | iodepth setting higher than 1, as it allows them to have |
| 1416 | IO in flight while verifies are running. |
| 1417 | |
| 1418 | verify_async_cpus=str Tell fio to set the given CPU affinity on the |
| 1419 | async IO verification threads. See cpus_allowed for the |
| 1420 | format used. |
| 1421 | |
| 1422 | verify_backlog=int Fio will normally verify the written contents of a |
| 1423 | job that utilizes verify once that job has completed. In |
| 1424 | other words, everything is written then everything is read |
| 1425 | back and verified. You may want to verify continually |
| 1426 | instead for a variety of reasons. Fio stores the meta data |
| 1427 | associated with an IO block in memory, so for large |
| 1428 | verify workloads, quite a bit of memory would be used up |
| 1429 | holding this meta data. If this option is enabled, fio |
| 1430 | will write only N blocks before verifying these blocks. |
| 1431 | |
| 1432 | verify_backlog_batch=int Control how many blocks fio will verify |
| 1433 | if verify_backlog is set. If not set, will default to |
| 1434 | the value of verify_backlog (meaning the entire queue |
| 1435 | is read back and verified). If verify_backlog_batch is |
| 1436 | less than verify_backlog then not all blocks will be verified, |
| 1437 | if verify_backlog_batch is larger than verify_backlog, some |
| 1438 | blocks will be verified more than once. |
| 1439 | |
| 1440 | verify_state_save=bool When a job exits during the write phase of a verify |
| 1441 | workload, save its current state. This allows fio to replay |
| 1442 | up until that point, if the verify state is loaded for the |
| 1443 | verify read phase. The format of the filename is, roughly, |
| 1444 | <type>-<jobname>-<jobindex>-verify.state. <type> is "local" |
| 1445 | for a local run, "sock" for a client/server socket connection, |
| 1446 | and "ip" (192.168.0.1, for instance) for a networked |
| 1447 | client/server connection. |
| 1448 | |
| 1449 | verify_state_load=bool If a verify termination trigger was used, fio stores |
| 1450 | the current write state of each thread. This can be used at |
| 1451 | verification time so that fio knows how far it should verify. |
| 1452 | Without this information, fio will run a full verification |
| 1453 | pass, according to the settings in the job file used. |
| 1454 | |
| 1455 | stonewall |
| 1456 | wait_for_previous Wait for preceding jobs in the job file to exit, before |
| 1457 | starting this one. Can be used to insert serialization |
| 1458 | points in the job file. A stone wall also implies starting |
| 1459 | a new reporting group. |
| 1460 | |
| 1461 | new_group Start a new reporting group. See: group_reporting. |
| 1462 | |
| 1463 | numjobs=int Create the specified number of clones of this job. May be |
| 1464 | used to setup a larger number of threads/processes doing |
| 1465 | the same thing. Each thread is reported separately; to see |
| 1466 | statistics for all clones as a whole, use group_reporting in |
| 1467 | conjunction with new_group. |
| 1468 | |
| 1469 | group_reporting It may sometimes be interesting to display statistics for |
| 1470 | groups of jobs as a whole instead of for each individual job. |
| 1471 | This is especially true if 'numjobs' is used; looking at |
| 1472 | individual thread/process output quickly becomes unwieldy. |
| 1473 | To see the final report per-group instead of per-job, use |
| 1474 | 'group_reporting'. Jobs in a file will be part of the same |
| 1475 | reporting group, unless if separated by a stonewall, or by |
| 1476 | using 'new_group'. |
| 1477 | |
| 1478 | thread fio defaults to forking jobs, however if this option is |
| 1479 | given, fio will use pthread_create(3) to create threads |
| 1480 | instead. |
| 1481 | |
| 1482 | zonesize=int Divide a file into zones of the specified size. See zoneskip. |
| 1483 | |
| 1484 | zoneskip=int Skip the specified number of bytes when zonesize data has |
| 1485 | been read. The two zone options can be used to only do |
| 1486 | io on zones of a file. |
| 1487 | |
| 1488 | write_iolog=str Write the issued io patterns to the specified file. See |
| 1489 | read_iolog. Specify a separate file for each job, otherwise |
| 1490 | the iologs will be interspersed and the file may be corrupt. |
| 1491 | |
| 1492 | read_iolog=str Open an iolog with the specified file name and replay the |
| 1493 | io patterns it contains. This can be used to store a |
| 1494 | workload and replay it sometime later. The iolog given |
| 1495 | may also be a blktrace binary file, which allows fio |
| 1496 | to replay a workload captured by blktrace. See blktrace |
| 1497 | for how to capture such logging data. For blktrace replay, |
| 1498 | the file needs to be turned into a blkparse binary data |
| 1499 | file first (blkparse <device> -o /dev/null -d file_for_fio.bin). |
| 1500 | |
| 1501 | replay_no_stall=int When replaying I/O with read_iolog the default behavior |
| 1502 | is to attempt to respect the time stamps within the log and |
| 1503 | replay them with the appropriate delay between IOPS. By |
| 1504 | setting this variable fio will not respect the timestamps and |
| 1505 | attempt to replay them as fast as possible while still |
| 1506 | respecting ordering. The result is the same I/O pattern to a |
| 1507 | given device, but different timings. |
| 1508 | |
| 1509 | replay_redirect=str While replaying I/O patterns using read_iolog the |
| 1510 | default behavior is to replay the IOPS onto the major/minor |
| 1511 | device that each IOP was recorded from. This is sometimes |
| 1512 | undesirable because on a different machine those major/minor |
| 1513 | numbers can map to a different device. Changing hardware on |
| 1514 | the same system can also result in a different major/minor |
| 1515 | mapping. Replay_redirect causes all IOPS to be replayed onto |
| 1516 | the single specified device regardless of the device it was |
| 1517 | recorded from. i.e. replay_redirect=/dev/sdc would cause all |
| 1518 | IO in the blktrace to be replayed onto /dev/sdc. This means |
| 1519 | multiple devices will be replayed onto a single, if the trace |
| 1520 | contains multiple devices. If you want multiple devices to be |
| 1521 | replayed concurrently to multiple redirected devices you must |
| 1522 | blkparse your trace into separate traces and replay them with |
| 1523 | independent fio invocations. Unfortuantely this also breaks |
| 1524 | the strict time ordering between multiple device accesses. |
| 1525 | |
| 1526 | replay_align=int Force alignment of IO offsets and lengths in a trace |
| 1527 | to this power of 2 value. |
| 1528 | |
| 1529 | replay_scale=int Scale sector offsets down by this factor when |
| 1530 | replaying traces. |
| 1531 | |
| 1532 | per_job_logs=bool If set, this generates bw/clat/iops log with per |
| 1533 | file private filenames. If not set, jobs with identical names |
| 1534 | will share the log filename. Default: true. |
| 1535 | |
| 1536 | write_bw_log=str If given, write a bandwidth log of the jobs in this job |
| 1537 | file. Can be used to store data of the bandwidth of the |
| 1538 | jobs in their lifetime. The included fio_generate_plots |
| 1539 | script uses gnuplot to turn these text files into nice |
| 1540 | graphs. See write_lat_log for behaviour of given |
| 1541 | filename. For this option, the suffix is _bw.x.log, where |
| 1542 | x is the index of the job (1..N, where N is the number of |
| 1543 | jobs). If 'per_job_logs' is false, then the filename will not |
| 1544 | include the job index. |
| 1545 | |
| 1546 | write_lat_log=str Same as write_bw_log, except that this option stores io |
| 1547 | submission, completion, and total latencies instead. If no |
| 1548 | filename is given with this option, the default filename of |
| 1549 | "jobname_type.log" is used. Even if the filename is given, |
| 1550 | fio will still append the type of log. So if one specifies |
| 1551 | |
| 1552 | write_lat_log=foo |
| 1553 | |
| 1554 | The actual log names will be foo_slat.x.log, foo_clat.x.log, |
| 1555 | and foo_lat.x.log, where x is the index of the job (1..N, |
| 1556 | where N is the number of jobs). This helps fio_generate_plot |
| 1557 | fine the logs automatically. If 'per_job_logs' is false, then |
| 1558 | the filename will not include the job index. |
| 1559 | |
| 1560 | |
| 1561 | write_iops_log=str Same as write_bw_log, but writes IOPS. If no filename is |
| 1562 | given with this option, the default filename of |
| 1563 | "jobname_type.x.log" is used,where x is the index of the job |
| 1564 | (1..N, where N is the number of jobs). Even if the filename |
| 1565 | is given, fio will still append the type of log. If |
| 1566 | 'per_job_logs' is false, then the filename will not include |
| 1567 | the job index. |
| 1568 | |
| 1569 | log_avg_msec=int By default, fio will log an entry in the iops, latency, |
| 1570 | or bw log for every IO that completes. When writing to the |
| 1571 | disk log, that can quickly grow to a very large size. Setting |
| 1572 | this option makes fio average the each log entry over the |
| 1573 | specified period of time, reducing the resolution of the log. |
| 1574 | See log_max as well. Defaults to 0, logging all entries. |
| 1575 | |
| 1576 | log_max=bool If log_avg_msec is set, fio logs the average over that window. |
| 1577 | If you instead want to log the maximum value, set this option |
| 1578 | to 1. Defaults to 0, meaning that averaged values are logged. |
| 1579 | . |
| 1580 | log_offset=int If this is set, the iolog options will include the byte |
| 1581 | offset for the IO entry as well as the other data values. |
| 1582 | |
| 1583 | log_compression=int If this is set, fio will compress the IO logs as |
| 1584 | it goes, to keep the memory footprint lower. When a log |
| 1585 | reaches the specified size, that chunk is removed and |
| 1586 | compressed in the background. Given that IO logs are |
| 1587 | fairly highly compressible, this yields a nice memory |
| 1588 | savings for longer runs. The downside is that the |
| 1589 | compression will consume some background CPU cycles, so |
| 1590 | it may impact the run. This, however, is also true if |
| 1591 | the logging ends up consuming most of the system memory. |
| 1592 | So pick your poison. The IO logs are saved normally at the |
| 1593 | end of a run, by decompressing the chunks and storing them |
| 1594 | in the specified log file. This feature depends on the |
| 1595 | availability of zlib. |
| 1596 | |
| 1597 | log_compression_cpus=str Define the set of CPUs that are allowed to |
| 1598 | handle online log compression for the IO jobs. This can |
| 1599 | provide better isolation between performance sensitive jobs, |
| 1600 | and background compression work. |
| 1601 | |
| 1602 | log_store_compressed=bool If set, fio will store the log files in a |
| 1603 | compressed format. They can be decompressed with fio, using |
| 1604 | the --inflate-log command line parameter. The files will be |
| 1605 | stored with a .fz suffix. |
| 1606 | |
| 1607 | block_error_percentiles=bool If set, record errors in trim block-sized |
| 1608 | units from writes and trims and output a histogram of |
| 1609 | how many trims it took to get to errors, and what kind |
| 1610 | of error was encountered. |
| 1611 | |
| 1612 | lockmem=int Pin down the specified amount of memory with mlock(2). Can |
| 1613 | potentially be used instead of removing memory or booting |
| 1614 | with less memory to simulate a smaller amount of memory. |
| 1615 | The amount specified is per worker. |
| 1616 | |
| 1617 | exec_prerun=str Before running this job, issue the command specified |
| 1618 | through system(3). Output is redirected in a file called |
| 1619 | jobname.prerun.txt. |
| 1620 | |
| 1621 | exec_postrun=str After the job completes, issue the command specified |
| 1622 | though system(3). Output is redirected in a file called |
| 1623 | jobname.postrun.txt. |
| 1624 | |
| 1625 | ioscheduler=str Attempt to switch the device hosting the file to the specified |
| 1626 | io scheduler before running. |
| 1627 | |
| 1628 | disk_util=bool Generate disk utilization statistics, if the platform |
| 1629 | supports it. Defaults to on. |
| 1630 | |
| 1631 | disable_lat=bool Disable measurements of total latency numbers. Useful |
| 1632 | only for cutting back the number of calls to gettimeofday, |
| 1633 | as that does impact performance at really high IOPS rates. |
| 1634 | Note that to really get rid of a large amount of these |
| 1635 | calls, this option must be used with disable_slat and |
| 1636 | disable_bw as well. |
| 1637 | |
| 1638 | disable_clat=bool Disable measurements of completion latency numbers. See |
| 1639 | disable_lat. |
| 1640 | |
| 1641 | disable_slat=bool Disable measurements of submission latency numbers. See |
| 1642 | disable_slat. |
| 1643 | |
| 1644 | disable_bw=bool Disable measurements of throughput/bandwidth numbers. See |
| 1645 | disable_lat. |
| 1646 | |
| 1647 | clat_percentiles=bool Enable the reporting of percentiles of |
| 1648 | completion latencies. |
| 1649 | |
| 1650 | percentile_list=float_list Overwrite the default list of percentiles |
| 1651 | for completion latencies and the block error histogram. |
| 1652 | Each number is a floating number in the range (0,100], |
| 1653 | and the maximum length of the list is 20. Use ':' |
| 1654 | to separate the numbers, and list the numbers in ascending |
| 1655 | order. For example, --percentile_list=99.5:99.9 will cause |
| 1656 | fio to report the values of completion latency below which |
| 1657 | 99.5% and 99.9% of the observed latencies fell, respectively. |
| 1658 | |
| 1659 | clocksource=str Use the given clocksource as the base of timing. The |
| 1660 | supported options are: |
| 1661 | |
| 1662 | gettimeofday gettimeofday(2) |
| 1663 | |
| 1664 | clock_gettime clock_gettime(2) |
| 1665 | |
| 1666 | cpu Internal CPU clock source |
| 1667 | |
| 1668 | cpu is the preferred clocksource if it is reliable, as it |
| 1669 | is very fast (and fio is heavy on time calls). Fio will |
| 1670 | automatically use this clocksource if it's supported and |
| 1671 | considered reliable on the system it is running on, unless |
| 1672 | another clocksource is specifically set. For x86/x86-64 CPUs, |
| 1673 | this means supporting TSC Invariant. |
| 1674 | |
| 1675 | gtod_reduce=bool Enable all of the gettimeofday() reducing options |
| 1676 | (disable_clat, disable_slat, disable_bw) plus reduce |
| 1677 | precision of the timeout somewhat to really shrink |
| 1678 | the gettimeofday() call count. With this option enabled, |
| 1679 | we only do about 0.4% of the gtod() calls we would have |
| 1680 | done if all time keeping was enabled. |
| 1681 | |
| 1682 | gtod_cpu=int Sometimes it's cheaper to dedicate a single thread of |
| 1683 | execution to just getting the current time. Fio (and |
| 1684 | databases, for instance) are very intensive on gettimeofday() |
| 1685 | calls. With this option, you can set one CPU aside for |
| 1686 | doing nothing but logging current time to a shared memory |
| 1687 | location. Then the other threads/processes that run IO |
| 1688 | workloads need only copy that segment, instead of entering |
| 1689 | the kernel with a gettimeofday() call. The CPU set aside |
| 1690 | for doing these time calls will be excluded from other |
| 1691 | uses. Fio will manually clear it from the CPU mask of other |
| 1692 | jobs. |
| 1693 | |
| 1694 | continue_on_error=str Normally fio will exit the job on the first observed |
| 1695 | failure. If this option is set, fio will continue the job when |
| 1696 | there is a 'non-fatal error' (EIO or EILSEQ) until the runtime |
| 1697 | is exceeded or the I/O size specified is completed. If this |
| 1698 | option is used, there are two more stats that are appended, |
| 1699 | the total error count and the first error. The error field |
| 1700 | given in the stats is the first error that was hit during the |
| 1701 | run. |
| 1702 | |
| 1703 | The allowed values are: |
| 1704 | |
| 1705 | none Exit on any IO or verify errors. |
| 1706 | |
| 1707 | read Continue on read errors, exit on all others. |
| 1708 | |
| 1709 | write Continue on write errors, exit on all others. |
| 1710 | |
| 1711 | io Continue on any IO error, exit on all others. |
| 1712 | |
| 1713 | verify Continue on verify errors, exit on all others. |
| 1714 | |
| 1715 | all Continue on all errors. |
| 1716 | |
| 1717 | 0 Backward-compatible alias for 'none'. |
| 1718 | |
| 1719 | 1 Backward-compatible alias for 'all'. |
| 1720 | |
| 1721 | ignore_error=str Sometimes you want to ignore some errors during test |
| 1722 | in that case you can specify error list for each error type. |
| 1723 | ignore_error=READ_ERR_LIST,WRITE_ERR_LIST,VERIFY_ERR_LIST |
| 1724 | errors for given error type is separated with ':'. Error |
| 1725 | may be symbol ('ENOSPC', 'ENOMEM') or integer. |
| 1726 | Example: |
| 1727 | ignore_error=EAGAIN,ENOSPC:122 |
| 1728 | This option will ignore EAGAIN from READ, and ENOSPC and |
| 1729 | 122(EDQUOT) from WRITE. |
| 1730 | |
| 1731 | error_dump=bool If set dump every error even if it is non fatal, true |
| 1732 | by default. If disabled only fatal error will be dumped |
| 1733 | |
| 1734 | cgroup=str Add job to this control group. If it doesn't exist, it will |
| 1735 | be created. The system must have a mounted cgroup blkio |
| 1736 | mount point for this to work. If your system doesn't have it |
| 1737 | mounted, you can do so with: |
| 1738 | |
| 1739 | # mount -t cgroup -o blkio none /cgroup |
| 1740 | |
| 1741 | cgroup_weight=int Set the weight of the cgroup to this value. See |
| 1742 | the documentation that comes with the kernel, allowed values |
| 1743 | are in the range of 100..1000. |
| 1744 | |
| 1745 | cgroup_nodelete=bool Normally fio will delete the cgroups it has created after |
| 1746 | the job completion. To override this behavior and to leave |
| 1747 | cgroups around after the job completion, set cgroup_nodelete=1. |
| 1748 | This can be useful if one wants to inspect various cgroup |
| 1749 | files after job completion. Default: false |
| 1750 | |
| 1751 | uid=int Instead of running as the invoking user, set the user ID to |
| 1752 | this value before the thread/process does any work. |
| 1753 | |
| 1754 | gid=int Set group ID, see uid. |
| 1755 | |
| 1756 | flow_id=int The ID of the flow. If not specified, it defaults to being a |
| 1757 | global flow. See flow. |
| 1758 | |
| 1759 | flow=int Weight in token-based flow control. If this value is used, then |
| 1760 | there is a 'flow counter' which is used to regulate the |
| 1761 | proportion of activity between two or more jobs. fio attempts |
| 1762 | to keep this flow counter near zero. The 'flow' parameter |
| 1763 | stands for how much should be added or subtracted to the flow |
| 1764 | counter on each iteration of the main I/O loop. That is, if |
| 1765 | one job has flow=8 and another job has flow=-1, then there |
| 1766 | will be a roughly 1:8 ratio in how much one runs vs the other. |
| 1767 | |
| 1768 | flow_watermark=int The maximum value that the absolute value of the flow |
| 1769 | counter is allowed to reach before the job must wait for a |
| 1770 | lower value of the counter. |
| 1771 | |
| 1772 | flow_sleep=int The period of time, in microseconds, to wait after the flow |
| 1773 | watermark has been exceeded before retrying operations |
| 1774 | |
| 1775 | In addition, there are some parameters which are only valid when a specific |
| 1776 | ioengine is in use. These are used identically to normal parameters, with the |
| 1777 | caveat that when used on the command line, they must come after the ioengine |
| 1778 | that defines them is selected. |
| 1779 | |
| 1780 | [libaio] userspace_reap Normally, with the libaio engine in use, fio will use |
| 1781 | the io_getevents system call to reap newly returned events. |
| 1782 | With this flag turned on, the AIO ring will be read directly |
| 1783 | from user-space to reap events. The reaping mode is only |
| 1784 | enabled when polling for a minimum of 0 events (eg when |
| 1785 | iodepth_batch_complete=0). |
| 1786 | |
| 1787 | [cpu] cpuload=int Attempt to use the specified percentage of CPU cycles. |
| 1788 | |
| 1789 | [cpu] cpuchunks=int Split the load into cycles of the given time. In |
| 1790 | microseconds. |
| 1791 | |
| 1792 | [cpu] exit_on_io_done=bool Detect when IO threads are done, then exit. |
| 1793 | |
| 1794 | [netsplice] hostname=str |
| 1795 | [net] hostname=str The host name or IP address to use for TCP or UDP based IO. |
| 1796 | If the job is a TCP listener or UDP reader, the hostname is not |
| 1797 | used and must be omitted unless it is a valid UDP multicast |
| 1798 | address. |
| 1799 | [libhdfs] namenode=str The host name or IP address of a HDFS cluster namenode to contact. |
| 1800 | |
| 1801 | [netsplice] port=int |
| 1802 | [net] port=int The TCP or UDP port to bind to or connect to. If this is used |
| 1803 | with numjobs to spawn multiple instances of the same job type, then this will |
| 1804 | be the starting port number since fio will use a range of ports. |
| 1805 | [libhdfs] port=int the listening port of the HFDS cluster namenode. |
| 1806 | |
| 1807 | [netsplice] interface=str |
| 1808 | [net] interface=str The IP address of the network interface used to send or |
| 1809 | receive UDP multicast |
| 1810 | |
| 1811 | [netsplice] ttl=int |
| 1812 | [net] ttl=int Time-to-live value for outgoing UDP multicast packets. |
| 1813 | Default: 1 |
| 1814 | |
| 1815 | [netsplice] nodelay=bool |
| 1816 | [net] nodelay=bool Set TCP_NODELAY on TCP connections. |
| 1817 | |
| 1818 | [netsplice] protocol=str |
| 1819 | [netsplice] proto=str |
| 1820 | [net] protocol=str |
| 1821 | [net] proto=str The network protocol to use. Accepted values are: |
| 1822 | |
| 1823 | tcp Transmission control protocol |
| 1824 | tcpv6 Transmission control protocol V6 |
| 1825 | udp User datagram protocol |
| 1826 | udpv6 User datagram protocol V6 |
| 1827 | unix UNIX domain socket |
| 1828 | |
| 1829 | When the protocol is TCP or UDP, the port must also be given, |
| 1830 | as well as the hostname if the job is a TCP listener or UDP |
| 1831 | reader. For unix sockets, the normal filename option should be |
| 1832 | used and the port is invalid. |
| 1833 | |
| 1834 | [net] listen For TCP network connections, tell fio to listen for incoming |
| 1835 | connections rather than initiating an outgoing connection. The |
| 1836 | hostname must be omitted if this option is used. |
| 1837 | |
| 1838 | [net] pingpong Normaly a network writer will just continue writing data, and |
| 1839 | a network reader will just consume packages. If pingpong=1 |
| 1840 | is set, a writer will send its normal payload to the reader, |
| 1841 | then wait for the reader to send the same payload back. This |
| 1842 | allows fio to measure network latencies. The submission |
| 1843 | and completion latencies then measure local time spent |
| 1844 | sending or receiving, and the completion latency measures |
| 1845 | how long it took for the other end to receive and send back. |
| 1846 | For UDP multicast traffic pingpong=1 should only be set for a |
| 1847 | single reader when multiple readers are listening to the same |
| 1848 | address. |
| 1849 | |
| 1850 | [net] window_size Set the desired socket buffer size for the connection. |
| 1851 | |
| 1852 | [net] mss Set the TCP maximum segment size (TCP_MAXSEG). |
| 1853 | |
| 1854 | [e4defrag] donorname=str |
| 1855 | File will be used as a block donor(swap extents between files) |
| 1856 | [e4defrag] inplace=int |
| 1857 | Configure donor file blocks allocation strategy |
| 1858 | 0(default): Preallocate donor's file on init |
| 1859 | 1 : allocate space immidietly inside defragment event, |
| 1860 | and free right after event |
| 1861 | |
| 1862 | [mtd] skip_bad=bool Skip operations against known bad blocks. |
| 1863 | |
| 1864 | [libhdfs] hdfsdirectory libhdfs will create chunk in this HDFS directory |
| 1865 | [libhdfs] chunck_size the size of the chunck to use for each file. |
| 1866 | |
| 1867 | |
| 1868 | 6.0 Interpreting the output |
| 1869 | --------------------------- |
| 1870 | |
| 1871 | fio spits out a lot of output. While running, fio will display the |
| 1872 | status of the jobs created. An example of that would be: |
| 1873 | |
| 1874 | Threads: 1: [_r] [24.8% done] [ 13509/ 8334 kb/s] [eta 00h:01m:31s] |
| 1875 | |
| 1876 | The characters inside the square brackets denote the current status of |
| 1877 | each thread. The possible values (in typical life cycle order) are: |
| 1878 | |
| 1879 | Idle Run |
| 1880 | ---- --- |
| 1881 | P Thread setup, but not started. |
| 1882 | C Thread created. |
| 1883 | I Thread initialized, waiting or generating necessary data. |
| 1884 | p Thread running pre-reading file(s). |
| 1885 | R Running, doing sequential reads. |
| 1886 | r Running, doing random reads. |
| 1887 | W Running, doing sequential writes. |
| 1888 | w Running, doing random writes. |
| 1889 | M Running, doing mixed sequential reads/writes. |
| 1890 | m Running, doing mixed random reads/writes. |
| 1891 | F Running, currently waiting for fsync() |
| 1892 | f Running, finishing up (writing IO logs, etc) |
| 1893 | V Running, doing verification of written data. |
| 1894 | E Thread exited, not reaped by main thread yet. |
| 1895 | _ Thread reaped, or |
| 1896 | X Thread reaped, exited with an error. |
| 1897 | K Thread reaped, exited due to signal. |
| 1898 | |
| 1899 | Fio will condense the thread string as not to take up more space on the |
| 1900 | command line as is needed. For instance, if you have 10 readers and 10 |
| 1901 | writers running, the output would look like this: |
| 1902 | |
| 1903 | Jobs: 20 (f=20): [R(10),W(10)] [4.0% done] [2103MB/0KB/0KB /s] [538K/0/0 iops] [eta 57m:36s] |
| 1904 | |
| 1905 | Fio will still maintain the ordering, though. So the above means that jobs |
| 1906 | 1..10 are readers, and 11..20 are writers. |
| 1907 | |
| 1908 | The other values are fairly self explanatory - number of threads |
| 1909 | currently running and doing io, rate of io since last check (read speed |
| 1910 | listed first, then write speed), and the estimated completion percentage |
| 1911 | and time for the running group. It's impossible to estimate runtime of |
| 1912 | the following groups (if any). Note that the string is displayed in order, |
| 1913 | so it's possible to tell which of the jobs are currently doing what. The |
| 1914 | first character is the first job defined in the job file, and so forth. |
| 1915 | |
| 1916 | When fio is done (or interrupted by ctrl-c), it will show the data for |
| 1917 | each thread, group of threads, and disks in that order. For each data |
| 1918 | direction, the output looks like: |
| 1919 | |
| 1920 | Client1 (g=0): err= 0: |
| 1921 | write: io= 32MB, bw= 666KB/s, iops=89 , runt= 50320msec |
| 1922 | slat (msec): min= 0, max= 136, avg= 0.03, stdev= 1.92 |
| 1923 | clat (msec): min= 0, max= 631, avg=48.50, stdev=86.82 |
| 1924 | bw (KB/s) : min= 0, max= 1196, per=51.00%, avg=664.02, stdev=681.68 |
| 1925 | cpu : usr=1.49%, sys=0.25%, ctx=7969, majf=0, minf=17 |
| 1926 | IO depths : 1=0.1%, 2=0.3%, 4=0.5%, 8=99.0%, 16=0.0%, 32=0.0%, >32=0.0% |
| 1927 | submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0% |
| 1928 | complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0% |
| 1929 | issued r/w: total=0/32768, short=0/0 |
| 1930 | lat (msec): 2=1.6%, 4=0.0%, 10=3.2%, 20=12.8%, 50=38.4%, 100=24.8%, |
| 1931 | lat (msec): 250=15.2%, 500=0.0%, 750=0.0%, 1000=0.0%, >=2048=0.0% |
| 1932 | |
| 1933 | The client number is printed, along with the group id and error of that |
| 1934 | thread. Below is the io statistics, here for writes. In the order listed, |
| 1935 | they denote: |
| 1936 | |
| 1937 | io= Number of megabytes io performed |
| 1938 | bw= Average bandwidth rate |
| 1939 | iops= Average IOs performed per second |
| 1940 | runt= The runtime of that thread |
| 1941 | slat= Submission latency (avg being the average, stdev being the |
| 1942 | standard deviation). This is the time it took to submit |
| 1943 | the io. For sync io, the slat is really the completion |
| 1944 | latency, since queue/complete is one operation there. This |
| 1945 | value can be in milliseconds or microseconds, fio will choose |
| 1946 | the most appropriate base and print that. In the example |
| 1947 | above, milliseconds is the best scale. Note: in --minimal mode |
| 1948 | latencies are always expressed in microseconds. |
| 1949 | clat= Completion latency. Same names as slat, this denotes the |
| 1950 | time from submission to completion of the io pieces. For |
| 1951 | sync io, clat will usually be equal (or very close) to 0, |
| 1952 | as the time from submit to complete is basically just |
| 1953 | CPU time (io has already been done, see slat explanation). |
| 1954 | bw= Bandwidth. Same names as the xlat stats, but also includes |
| 1955 | an approximate percentage of total aggregate bandwidth |
| 1956 | this thread received in this group. This last value is |
| 1957 | only really useful if the threads in this group are on the |
| 1958 | same disk, since they are then competing for disk access. |
| 1959 | cpu= CPU usage. User and system time, along with the number |
| 1960 | of context switches this thread went through, usage of |
| 1961 | system and user time, and finally the number of major |
| 1962 | and minor page faults. |
| 1963 | IO depths= The distribution of io depths over the job life time. The |
| 1964 | numbers are divided into powers of 2, so for example the |
| 1965 | 16= entries includes depths up to that value but higher |
| 1966 | than the previous entry. In other words, it covers the |
| 1967 | range from 16 to 31. |
| 1968 | IO submit= How many pieces of IO were submitting in a single submit |
| 1969 | call. Each entry denotes that amount and below, until |
| 1970 | the previous entry - eg, 8=100% mean that we submitted |
| 1971 | anywhere in between 5-8 ios per submit call. |
| 1972 | IO complete= Like the above submit number, but for completions instead. |
| 1973 | IO issued= The number of read/write requests issued, and how many |
| 1974 | of them were short. |
| 1975 | IO latencies= The distribution of IO completion latencies. This is the |
| 1976 | time from when IO leaves fio and when it gets completed. |
| 1977 | The numbers follow the same pattern as the IO depths, |
| 1978 | meaning that 2=1.6% means that 1.6% of the IO completed |
| 1979 | within 2 msecs, 20=12.8% means that 12.8% of the IO |
| 1980 | took more than 10 msecs, but less than (or equal to) 20 msecs. |
| 1981 | |
| 1982 | After each client has been listed, the group statistics are printed. They |
| 1983 | will look like this: |
| 1984 | |
| 1985 | Run status group 0 (all jobs): |
| 1986 | READ: io=64MB, aggrb=22178, minb=11355, maxb=11814, mint=2840msec, maxt=2955msec |
| 1987 | WRITE: io=64MB, aggrb=1302, minb=666, maxb=669, mint=50093msec, maxt=50320msec |
| 1988 | |
| 1989 | For each data direction, it prints: |
| 1990 | |
| 1991 | io= Number of megabytes io performed. |
| 1992 | aggrb= Aggregate bandwidth of threads in this group. |
| 1993 | minb= The minimum average bandwidth a thread saw. |
| 1994 | maxb= The maximum average bandwidth a thread saw. |
| 1995 | mint= The smallest runtime of the threads in that group. |
| 1996 | maxt= The longest runtime of the threads in that group. |
| 1997 | |
| 1998 | And finally, the disk statistics are printed. They will look like this: |
| 1999 | |
| 2000 | Disk stats (read/write): |
| 2001 | sda: ios=16398/16511, merge=30/162, ticks=6853/819634, in_queue=826487, util=100.00% |
| 2002 | |
| 2003 | Each value is printed for both reads and writes, with reads first. The |
| 2004 | numbers denote: |
| 2005 | |
| 2006 | ios= Number of ios performed by all groups. |
| 2007 | merge= Number of merges io the io scheduler. |
| 2008 | ticks= Number of ticks we kept the disk busy. |
| 2009 | io_queue= Total time spent in the disk queue. |
| 2010 | util= The disk utilization. A value of 100% means we kept the disk |
| 2011 | busy constantly, 50% would be a disk idling half of the time. |
| 2012 | |
| 2013 | It is also possible to get fio to dump the current output while it is |
| 2014 | running, without terminating the job. To do that, send fio the USR1 signal. |
| 2015 | You can also get regularly timed dumps by using the --status-interval |
| 2016 | parameter, or by creating a file in /tmp named fio-dump-status. If fio |
| 2017 | sees this file, it will unlink it and dump the current output status. |
| 2018 | |
| 2019 | |
| 2020 | 7.0 Terse output |
| 2021 | ---------------- |
| 2022 | |
| 2023 | For scripted usage where you typically want to generate tables or graphs |
| 2024 | of the results, fio can output the results in a semicolon separated format. |
| 2025 | The format is one long line of values, such as: |
| 2026 | |
| 2027 | 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% |
| 2028 | A description of this job goes here. |
| 2029 | |
| 2030 | The job description (if provided) follows on a second line. |
| 2031 | |
| 2032 | To enable terse output, use the --minimal command line option. The first |
| 2033 | value is the version of the terse output format. If the output has to |
| 2034 | be changed for some reason, this number will be incremented by 1 to |
| 2035 | signify that change. |
| 2036 | |
| 2037 | Split up, the format is as follows: |
| 2038 | |
| 2039 | terse version, fio version, jobname, groupid, error |
| 2040 | READ status: |
| 2041 | Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec) |
| 2042 | Submission latency: min, max, mean, stdev (usec) |
| 2043 | Completion latency: min, max, mean, stdev (usec) |
| 2044 | Completion latency percentiles: 20 fields (see below) |
| 2045 | Total latency: min, max, mean, stdev (usec) |
| 2046 | Bw (KB/s): min, max, aggregate percentage of total, mean, stdev |
| 2047 | WRITE status: |
| 2048 | Total IO (KB), bandwidth (KB/sec), IOPS, runtime (msec) |
| 2049 | Submission latency: min, max, mean, stdev (usec) |
| 2050 | Completion latency: min, max, mean, stdev(usec) |
| 2051 | Completion latency percentiles: 20 fields (see below) |
| 2052 | Total latency: min, max, mean, stdev (usec) |
| 2053 | Bw (KB/s): min, max, aggregate percentage of total, mean, stdev |
| 2054 | CPU usage: user, system, context switches, major faults, minor faults |
| 2055 | IO depths: <=1, 2, 4, 8, 16, 32, >=64 |
| 2056 | IO latencies microseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000 |
| 2057 | IO latencies milliseconds: <=2, 4, 10, 20, 50, 100, 250, 500, 750, 1000, 2000, >=2000 |
| 2058 | Disk utilization: Disk name, Read ios, write ios, |
| 2059 | Read merges, write merges, |
| 2060 | Read ticks, write ticks, |
| 2061 | Time spent in queue, disk utilization percentage |
| 2062 | Additional Info (dependent on continue_on_error, default off): total # errors, first error code |
| 2063 | |
| 2064 | Additional Info (dependent on description being set): Text description |
| 2065 | |
| 2066 | Completion latency percentiles can be a grouping of up to 20 sets, so |
| 2067 | for the terse output fio writes all of them. Each field will look like this: |
| 2068 | |
| 2069 | 1.00%=6112 |
| 2070 | |
| 2071 | which is the Xth percentile, and the usec latency associated with it. |
| 2072 | |
| 2073 | For disk utilization, all disks used by fio are shown. So for each disk |
| 2074 | there will be a disk utilization section. |
| 2075 | |
| 2076 | |
| 2077 | 8.0 Trace file format |
| 2078 | --------------------- |
| 2079 | There are two trace file format that you can encounter. The older (v1) format |
| 2080 | is unsupported since version 1.20-rc3 (March 2008). It will still be described |
| 2081 | below in case that you get an old trace and want to understand it. |
| 2082 | |
| 2083 | In any case the trace is a simple text file with a single action per line. |
| 2084 | |
| 2085 | |
| 2086 | 8.1 Trace file format v1 |
| 2087 | ------------------------ |
| 2088 | Each line represents a single io action in the following format: |
| 2089 | |
| 2090 | rw, offset, length |
| 2091 | |
| 2092 | where rw=0/1 for read/write, and the offset and length entries being in bytes. |
| 2093 | |
| 2094 | This format is not supported in Fio versions => 1.20-rc3. |
| 2095 | |
| 2096 | |
| 2097 | 8.2 Trace file format v2 |
| 2098 | ------------------------ |
| 2099 | The second version of the trace file format was added in Fio version 1.17. |
| 2100 | It allows to access more then one file per trace and has a bigger set of |
| 2101 | possible file actions. |
| 2102 | |
| 2103 | The first line of the trace file has to be: |
| 2104 | |
| 2105 | fio version 2 iolog |
| 2106 | |
| 2107 | Following this can be lines in two different formats, which are described below. |
| 2108 | |
| 2109 | The file management format: |
| 2110 | |
| 2111 | filename action |
| 2112 | |
| 2113 | The filename is given as an absolute path. The action can be one of these: |
| 2114 | |
| 2115 | add Add the given filename to the trace |
| 2116 | open Open the file with the given filename. The filename has to have |
| 2117 | been added with the add action before. |
| 2118 | close Close the file with the given filename. The file has to have been |
| 2119 | opened before. |
| 2120 | |
| 2121 | |
| 2122 | The file io action format: |
| 2123 | |
| 2124 | filename action offset length |
| 2125 | |
| 2126 | The filename is given as an absolute path, and has to have been added and opened |
| 2127 | before it can be used with this format. The offset and length are given in |
| 2128 | bytes. The action can be one of these: |
| 2129 | |
| 2130 | wait Wait for 'offset' microseconds. Everything below 100 is discarded. |
| 2131 | The time is relative to the previous wait statement. |
| 2132 | read Read 'length' bytes beginning from 'offset' |
| 2133 | write Write 'length' bytes beginning from 'offset' |
| 2134 | sync fsync() the file |
| 2135 | datasync fdatasync() the file |
| 2136 | trim trim the given file from the given 'offset' for 'length' bytes |
| 2137 | |
| 2138 | |
| 2139 | 9.0 CPU idleness profiling |
| 2140 | -------------------------- |
| 2141 | In some cases, we want to understand CPU overhead in a test. For example, |
| 2142 | we test patches for the specific goodness of whether they reduce CPU usage. |
| 2143 | fio implements a balloon approach to create a thread per CPU that runs at |
| 2144 | idle priority, meaning that it only runs when nobody else needs the cpu. |
| 2145 | By measuring the amount of work completed by the thread, idleness of each |
| 2146 | CPU can be derived accordingly. |
| 2147 | |
| 2148 | An unit work is defined as touching a full page of unsigned characters. Mean |
| 2149 | and standard deviation of time to complete an unit work is reported in "unit |
| 2150 | work" section. Options can be chosen to report detailed percpu idleness or |
| 2151 | overall system idleness by aggregating percpu stats. |
| 2152 | |
| 2153 | |
| 2154 | 10.0 Verification and triggers |
| 2155 | ------------------------------ |
| 2156 | Fio is usually run in one of two ways, when data verification is done. The |
| 2157 | first is a normal write job of some sort with verify enabled. When the |
| 2158 | write phase has completed, fio switches to reads and verifies everything |
| 2159 | it wrote. The second model is running just the write phase, and then later |
| 2160 | on running the same job (but with reads instead of writes) to repeat the |
| 2161 | same IO patterns and verify the contents. Both of these methods depend |
| 2162 | on the write phase being completed, as fio otherwise has no idea how much |
| 2163 | data was written. |
| 2164 | |
| 2165 | With verification triggers, fio supports dumping the current write state |
| 2166 | to local files. Then a subsequent read verify workload can load this state |
| 2167 | and know exactly where to stop. This is useful for testing cases where |
| 2168 | power is cut to a server in a managed fashion, for instance. |
| 2169 | |
| 2170 | A verification trigger consists of two things: |
| 2171 | |
| 2172 | 1) Storing the write state of each job |
| 2173 | 2) Executing a trigger command |
| 2174 | |
| 2175 | The write state is relatively small, on the order of hundreds of bytes |
| 2176 | to single kilobytes. It contains information on the number of completions |
| 2177 | done, the last X completions, etc. |
| 2178 | |
| 2179 | A trigger is invoked either through creation ('touch') of a specified |
| 2180 | file in the system, or through a timeout setting. If fio is run with |
| 2181 | --trigger-file=/tmp/trigger-file, then it will continually check for |
| 2182 | the existence of /tmp/trigger-file. When it sees this file, it will |
| 2183 | fire off the trigger (thus saving state, and executing the trigger |
| 2184 | command). |
| 2185 | |
| 2186 | For client/server runs, there's both a local and remote trigger. If |
| 2187 | fio is running as a server backend, it will send the job states back |
| 2188 | to the client for safe storage, then execute the remote trigger, if |
| 2189 | specified. If a local trigger is specified, the server will still send |
| 2190 | back the write state, but the client will then execute the trigger. |
| 2191 | |
| 2192 | 10.1 Verification trigger example |
| 2193 | --------------------------------- |
| 2194 | Lets say we want to run a powercut test on the remote machine 'server'. |
| 2195 | Our write workload is in write-test.fio. We want to cut power to 'server' |
| 2196 | at some point during the run, and we'll run this test from the safety |
| 2197 | or our local machine, 'localbox'. On the server, we'll start the fio |
| 2198 | backend normally: |
| 2199 | |
| 2200 | server# fio --server |
| 2201 | |
| 2202 | and on the client, we'll fire off the workload: |
| 2203 | |
| 2204 | localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger-remote="bash -c \"echo b > /proc/sysrq-triger\"" |
| 2205 | |
| 2206 | We set /tmp/my-trigger as the trigger file, and we tell fio to execute |
| 2207 | |
| 2208 | echo b > /proc/sysrq-trigger |
| 2209 | |
| 2210 | on the server once it has received the trigger and sent us the write |
| 2211 | state. This will work, but it's not _really_ cutting power to the server, |
| 2212 | it's merely abruptly rebooting it. If we have a remote way of cutting |
| 2213 | power to the server through IPMI or similar, we could do that through |
| 2214 | a local trigger command instead. Lets assume we have a script that does |
| 2215 | IPMI reboot of a given hostname, ipmi-reboot. On localbox, we could |
| 2216 | then have run fio with a local trigger instead: |
| 2217 | |
| 2218 | localbox$ fio --client=server --trigger-file=/tmp/my-trigger --trigger="ipmi-reboot server" |
| 2219 | |
| 2220 | For this case, fio would wait for the server to send us the write state, |
| 2221 | then execute 'ipmi-reboot server' when that happened. |
| 2222 | |
| 2223 | 10.2 Loading verify state |
| 2224 | ------------------------- |
| 2225 | To load store write state, read verification job file must contain |
| 2226 | the verify_state_load option. If that is set, fio will load the previously |
| 2227 | stored state. For a local fio run this is done by loading the files directly, |
| 2228 | and on a client/server run, the server backend will ask the client to send |
| 2229 | the files over and load them from there. |