FIO engine to test "file server" access pattern
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Vitaliy Filippov c49ecf80f9 Add compatibility note 3 years ago
COPYING Initial commit 3 years ago
Makefile It's better to build it with C, not C++ 3 years ago Add compatibility note 3 years ago
fio_fileserver.c It's better to build it with C, not C++ 3 years ago


This is a loadable engine for fio to test something similar to the file server access pattern where a number of files is sharded over several levels of subdirectories.


  • Clone this repository
  • Clone or symlink the source code directory for your exact fio version in ./fio subdirectory
  • Install g++
  • Run make


fio 3.15 or later.


fio -name=test -ioengine=./ -directory=/home/bench \
    -chunk_size=256k -size=10G -direct=1 [-dir_levels=2] [-subdirs_per_dir=64] \
    -bs=256k -rw=randwrite [-fsync_on_close=1] [-sync=1] [-numjobs=16 -group_reporting]


  • direct=1 is optional, but without it you'll be benchmarking the page cache
  • bs=256k must be less or equal to chunk_size
  • fsync=N is supported, but it only fsyncs a random file
  • reads return error when encountering a non-existing file
  • You can also use various distribution parameters (zipf, etc)
  • I/O is done using synchronous syscalls, thus iodepth is not supported

Because of the synchronous syscalls it's slightly tricky to use parallelism correctly. So, for example, to fill a directory with 262144 4k files, run:

fio -thread -name=test -direct=1 -ioengine=./ -fsync_on_close=1 \
    -directory=/home/bench -chunk_size=4k -size=1G -bs=4k -rw=write:508k \
    -numjobs=128 -offset_increment=4k -group_reporting

License & Author

  • Author: Vitaliy Filippov vitalif[at], 2020
  • License: GNU GPL v2.0 or later version