使用fio测试磁盘I/O性能

简介：

fio是测试IOPS的非常好的工具，用来对硬件进行压力测试和验证，支持13种不同的I/O引擎，包括:sync,mmap, libaio, posixaio, SG v3, splice, null, network, syslet, guasi, solarisaio 等等。

安装fio：

官网网站：http://brick.kernel.dk/snaps/

wget http://brick.kernel.dk/snaps/fio-2.2.10.tar.gz (从官网下载fio的tar包)
tar -zxvf fio-2.2..tar.gz　　　　　　　　　　　　　　　　　（解压）
cd fio-2.2.
make && make install　　　　　　　　　　　　　　　　　　（安装）
yum install libaio-devel　　　　　　　　　　　　　　　　　（为了支持同步I/O）

fio分随机读，随机写，顺序读，顺序写，混合随机读写模式。

fio参数：

filename=/dev/sdb　　　　#测试文件(设备)的名称，通常选择需要测试的data目录（可以通过冒号分割同时指定多个文件，如filename=/dev/sda:/dev/sdb。）

direct=　　　　　　　      #测试过程绕过机器自带的buffer（缓冲区），使测试结果更真实

rw=randwrite　　　    　　·　　　　 #测试随机写的I/O

rw=randread   　　　　　　　　　 #测试随机读的I/O

rw=randrw　　　　　　　　　　　 #测试随机读写的I/O

bs=4k　　　　　　　　　　　　　 #单次I/O的块文件大小为4K

size=20G　　　　　　　　　　 #本次测试的文件大小为20G，以每次4K的I/O进行测试

numjobs=　　　　　　　　　#本次的测试线程为30个

runtime=　　　　　　　#测试时间为600秒，如果不写，则一直将20G文件分每次4K写完为止

ioengine=psync　　　　　 #I/O引擎使用psync方式（I/O引擎，现在fio支持19种ioengine。默认值是sync同步阻塞I/O，libaio是Linux的native异步I/O。关于同步异步，阻塞和非阻塞模型可以参考文章“使用异步 I/O 大大提高应用程序的性能”。
http://www.ibm.com/developerworks/cn/linux/l-async/）

group_reporting　　　　　#关于显示结果的，汇总每个进程的信息

name　　　　　　　　　　#指定job的名字，在命令行中表示新启动一个job。

iodepth　　　　　　　　　#如果ioengine采用异步方式，该参数表示一批提交保持的io单元数。（该参数可参考文章“Fio压测工具和io队列深度理解和误区”。http://blog.yufeng.info/archives/2104）

thread　　　　　　　　　 #线程

fio实例：

随机读4K：

fio -filename=/dev/sdb:/dev/sdc:/dev/sdd -direct= -iodepth= -thread -rw=randread -ioengine=libaio -bs=4k -size=20G -numjobs= -runtime= -group_reporting -name=mytest

随机写4K：

fio -filename=/dev/sdb:/dev/sdc:/dev/sdd -direct= -iodepth= -thread -rw=randwrite -ioengine=libaio -bs=4k -size=20G -numjobs= -runtime= -group_reporting -name=mytest

顺序读1M：

fio -filename=/dev/sdb:/dev/sdc:/dev/sdd -direct= -iodepth= -thread -rw=read -ioengine=libaio -bs=1M -size=20G -numjobs= -runtime= -group_reporting -name=mytest

顺序写1M：

fio -filename=/dev/sdb:/dev/sdc:/dev/sdd -direct= -iodepth= -thread -rw=write -ioengine=libaio -bs=1M -size=20G -numjobs= -runtime= -group_reporting -name=mytest

混合随机读写：

fio -filename=/dev/sdb:/dev/sdc:/dev/sdd -direct= -iodepth  -thread -rw=randrw -rwmixread= -ioengine=psync -bs=16k -size=200G -numjobs= -runtime= -group_reporting -name=mytest -ioscheduler=noop

实例测试：

注：主要关注结果用橙色标出

4k随机读：

[root@localhost ~]# fio -filename=/dev/sda:/dev/sdb:/dev/sdc -direct= -iodepth  -thread -rw=randread -ioengine=libaio -bs=4k -size=20G -numjobs= -runtime= -group_reporting -name=mytest
mytest: (g=): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=
...
fio-2.21
Starting  threads
Jobs:  (f=): [r()][100.0%][r=.2MiB/s,w=0KiB/s][r=,w= IOPS][eta 00m:00s]
mytest: (groupid=, jobs=): err= : pid=: Thu Jun  ::
   read: IOPS=, BW=.8MiB/s (.3MB/s)(1077MiB/100033msec)
    slat (usec): min=, max=, avg=21.55, stdev= 9.74
    clat (usec): min=, max=, avg=10853.67, stdev=12909.46
     lat (usec): min=, max=, avg=10875.94, stdev=12909.57
    clat percentiles (usec):
     |  .00th=[ ],  .00th=[ ], .00th=[ ], .00th=[ ],
     | .00th=[ ], .00th=[ ], .00th=[ ], .00th=[],
     | .00th=[], .00th=[], .00th=[], .00th=[],
     | .00th=[], .50th=[], .90th=[], .95th=[],
     | .99th=[]
   bw (  KiB/s): min=   , max=  , per=0.00%, avg=369.92, stdev=100.81
    lat (usec) : =0.11%, =0.12%, =0.27%
    lat (msec) : =2.59%, =9.55%, =43.26%, =37.74%, =5.51%
    lat (msec) : =0.63%, =0.20%, =0.01%, =0.01%
  cpu          : usr=0.09%, sys=0.32%, ctx=, majf=, minf=
  IO depths    : =100.0%, =0.0%, =0.0%, =0.0%, =0.0%, =0.0%, >==0.0%
     submit    : =0.0%, =100.0%, =0.0%, =0.0%, =0.0%, =0.0%, >==0.0%
     complete  : =0.0%, =100.0%, =0.0%, =0.0%, =0.0%, =0.0%, >==0.0%
     issued rwt: total=,,, short=,,, dropped=,,
     latency   : target=, window=, percentile=100.00%, depth=

Run status group  (all jobs):
   READ: bw=.8MiB/s (.3MB/s), .8MiB/s-.8MiB/s (.3MB/s-.3MB/s), io=1077MiB (1129MB), run=-100033msec

Disk stats (read/write):
  sda: ios=/, merge=/, ticks=/, in_queue=, util=92.52%
  sdb: ios=/, merge=/, ticks=/, in_queue=, util=84.34%
  sdc: ios=/, merge=/, ticks=/, in_queue=, util=91.21%

混合随机读写：

[root@localhost ~]# fio -filename=/dev/sda:/dev/sdb:/dev/sdc -direct= -iodepth  -thread -rw=randrw -ioengine=libaio -bs=4k -size=20G -numjobs= -runtime= -group_reporting -name=mytest
mytest: (g=): rw=randrw, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=
...
fio-2.21
Starting  threads
Jobs:  (f=): [m()][100.0%][r=404KiB/s,w=464KiB/s][r=,w= IOPS][eta 00m:00s]
mytest: (groupid=, jobs=): err= : pid=: Thu Jun  ::
   read: IOPS=, BW=345KiB/s (353kB/s)(.7MiB/100117msec)
    slat (usec): min=, max=, avg=30.70, stdev=11.21
    clat (usec): min=, max=, avg=50031.54, stdev=79240.40
     lat (usec): min=, max=, avg=50063.31, stdev=79240.09
    clat percentiles (usec):
     |  .00th=[  ],  .00th=[ ], .00th=[ ], .00th=[ ],
     | .00th=[ ], .00th=[ ], .00th=[], .00th=[],
     | .00th=[], .00th=[], .00th=[], .00th=[],
     | .00th=[], .50th=[], .90th=[], .95th=[],
     | .99th=[]
   bw (  KiB/s): min=    , max=   , per=0.01%, avg=18.68, stdev=12.27
  write: IOPS=, BW=344KiB/s (352kB/s)(.6MiB/100117msec)
    slat (usec): min=, max=, avg=34.27, stdev=15.34
    clat (msec): min=, max=, avg=297.95, stdev=232.05
     lat (msec): min=, max=, avg=297.99, stdev=232.05
    clat percentiles (msec):
     |  .00th=[    ],  .00th=[   ], .00th=[   ], .00th=[  ],
     | .00th=[  ], .00th=[  ], .00th=[  ], .00th=[  ],
     | .00th=[  ], .00th=[  ], .00th=[  ], .00th=[  ],
     | .00th=[ ], .50th=[ ], .90th=[ ], .95th=[ ],
     | .99th=[ ]
   bw (  KiB/s): min=    , max=   , per=0.00%, avg=13.50, stdev= 6.61
    lat (usec) : =0.02%, =0.48%, =0.67%
    lat (msec) : =5.84%, =5.92%, =10.74%, =7.24%, =9.34%
    lat (msec) : =10.07%, =23.98%, =17.14%, =5.98%, =1.94%
    lat (msec) : =0.64%
  cpu          : usr=0.01%, sys=0.03%, ctx=, majf=, minf=
  IO depths    : =100.0%, =0.0%, =0.0%, =0.0%, =0.0%, =0.0%, >==0.0%
     submit    : =0.0%, =100.0%, =0.0%, =0.0%, =0.0%, =0.0%, >==0.0%
     complete  : =0.0%, =100.0%, =0.0%, =0.0%, =0.0%, =0.0%, >==0.0%
     issued rwt: total=,,, short=,,, dropped=,,
     latency   : target=, window=, percentile=100.00%, depth=

Run status group  (all jobs):
   READ: bw=345KiB/s (353kB/s), 345KiB/s-345KiB/s (353kB/s-353kB/s), io=.7MiB (.4MB), run=-100117msec
  WRITE: bw=344KiB/s (352kB/s), 344KiB/s-344KiB/s (352kB/s-352kB/s), io=.6MiB (.3MB), run=-100117msec

Disk stats (read/write):
  sda: ios=/, merge=/, ticks=/, in_queue=, util=100.00%
  sdb: ios=/, merge=/, ticks=/, in_queue=, util=99.66%
  sdc: ios=/, merge=/, ticks=/, in_queue=, util=99.90%

硬盘性能指标：

顺序读写 （吞吐量，常用单位为MB/s）：文件在硬盘上存储位置是连续的。

适用场景：大文件拷贝（比如视频音乐）。速度即使很高，对数据库性能也没有参考价值。

4K随机读写 （IOPS，常用单位为次）：在硬盘上随机位置读写数据，每次4KB。

适用场景：操作系统运行、软件运行、数据库。