当前位置：网站首页>FIO test hard disk performance

FIO test hard disk performance

2022-07-20 10:19:00 【The first code kills the immortal】

FIO It is a tool for testing disk performance , For stress testing and verification of hardware .
Use FIO when , It is recommended to use with libaio Of I/O Test the engine .

Software installation

Compilation and installation

fio The download address of each version is as follows ：

https://github.com/axboe/fio/releases

With “fio-2.1.10” Version as an example , Download the source code , And compile and install .

# Installation dependency 
yum install libaio libaio-devel -y
# download Fio Source code .
wget https://github.com/axboe/fio/archive/fio-2.1.10.tar.gz
#  decompression Fio Source code .
tar -zxvf fio-2.1.10.tar.gz
# Compile and install Fio.
cd fio-fio-2.1.10
make
make install
#  Check the installation Fio Version number .
fio --version
# The echo message is as follows , be Fio Installation successful .
fio-2.1.10

notes ： You may need to exit the current shell, Log in again

yum install

If the version is not required , Can directly yum install ：

yum install libaio libaio-devel -y
yum install -y fio

Measures

IOPS： Read every second / Write times , The unit is secondary （ Count ）. The underlying driver type of the storage device determines the different IOPS.
throughput ： The amount of read and write data per second , Unit is MB/s.
Time delay ：I/O The time between sending the operation and receiving the confirmation , The unit is in seconds .

Test suggestions

It is suggested to do it in the free 、 On the hard disk without saving important data FIO test , And remake the file system of the tested hard disk after the test .
When testing the performance of hard disk , It is recommended to test the raw data disk directly （ Such as /dev/vdb）.
When testing file system performance , It is recommended to specify specific document tests （ Such as /data/file）.

Description of common parameters

The test formulas of different scenarios are basically the same , Only rw、iodepth and bs（block size） The difference between the three parameters . for example , Each workload fits best iodepth Different , It depends on your specific application for IOPS And the sensitivity of delay .

Parameter name	explain	Examples
bs	Block size per request . Values include 4k,8k,16k etc. .	4k
ioengine	I/O engine . Recommended Linux The asynchronous I/O engine .	libaio
iodepth	Requested I/O Queue depth .	1
direct	Appoint direct Pattern . True（1） To designate O_DIRECT identifier , Ignore I/O cache , Data direct writing . False（0） Means not to specify O_DIRECT identifier . The default is True（1）.	1
rw	Read write mode . Values include sequential reads （read）、 Sequential writing （write）、 random block read （randread）、 Write at random （randwrite）、 Mixed random read and write （randrw） And mixed sequential reading and writing （rw,readwrite）.	read
time_based	Specify the time mode . There is no need to set the parameter value , as long as FIO Run based on time .	N/A
runtime	Specify test duration , namely FIO Run time .	300
refill_buffers	FIO Will be repopulated on each submission I/O buffer . The default is to fill and reuse the data only initially .	N/A
norandommap	Random I/O when ,FIO Each block of the file will be overwritten . If this parameter is given , Then the new offset will be selected without viewing I/O Historical record .	N/A
randrepeat	Whether the random sequence is repeatable ,True（1） Indicates that the random sequence is repeatable ,False（0） Indicates that the random sequence is not repeatable . The default is True（1）.	0
group_reporting	Multiple job Concurrent , Print the whole group The statistical value of .	N/A
name	job The name of .	fio-read
size	I/O Address space tested . When not specified , Will test the whole . If it is a file test , If you don't specify the size , Will report a mistake .	100G
filename	Test object , That is, the name of the disk device to be tested .	/dev/sdb
numjobs	Number of test processes	16

The test case

Bare disk iops test

Write iops

fio -numjobs=16 -bs=4k -ioengine=libaio -iodepth=32 -direct=1 -rw=randwrite -time_based -runtime=300  -refill_buffers -norandommap -randrepeat=0 -group_reporting -name=fio-randwrite-iops -filename=/dev/vdb

Write iops Pay attention to the following output
Insert picture description here

read iops

fio -numjobs=16 -bs=4k -ioengine=libaio -iodepth=32 -direct=1 -rw=randread -time_based -runtime=300  -refill_buffers -norandommap -randrepeat=0 -group_reporting -name=fio-randread-iops -filename=/dev/vdb

read iops Pay attention to the following output ：
Insert picture description here

Bare disk bandwidth test （ Sequential reading and writing ）

Sequential writing

fio --numjobs=16 -bs=128k -ioengine=libaio -iodepth=32 -direct=1 -rw=write -time_based -runtime=300  -refill_buffers -norandommap -randrepeat=0 -group_reporting -name=fio-write-throughput -filename=/dev/vdb

Sequential reading

fio --numjobs=16 -bs=128k -ioengine=libaio -iodepth=32 -direct=1 -rw=read -time_based -runtime=300  -refill_buffers -norandommap -randrepeat=0 -group_reporting -name=fio-read-throughput -filename=/dev/vdb

Raw disk delay

Write delay

fio --numjobs=16 -bs=4k -ioengine=libaio -iodepth=1 -direct=1 -rw=randwrite -time_based -runtime=300  -refill_buffers -norandommap -randrepeat=0 -group_reporting -name=fio-randwrite-lat -filename=/dev/vdb

Read delay

fio --numjobs=16 -bs=4k -ioengine=libaio -iodepth=1 -direct=1 -rw=randread -time_based -runtime=300  -refill_buffers -norandommap -randrepeat=0 -group_reporting -name=fio-randread-lat -filename=/dev/vdb