1. Performance metrics for cloud hard drives
The performance indicators of cloud hard drives are generally measured by the following indicators
- IOPS: The number of reads and writes per second, in times (count). The underlying drive type of the storage device determines the
total IOPS of different IOPS: the total number of I/O operations performed
per second . Random read IOPS: the average number of random read I/O operations
specified per second . Random write IOPS. Random write specified per second. The average number of I/O operations
Sequential read IOPS The average number of sequential read I/O operations
specified per second Sequential write IOPS The average number of sequential write I/O operations specified per second - Throughput: the amount of read and write data per second, the unit is the amount
of data that can be successfully transmitted per unit time of the market value of MB/S throughput.
If you need to deploy a large number of sequential read and write applications, typical scenarios such as Hadoop offline computing services, you need to pay attention to throughput -
Latency: The elapsed time from the sending time of the IO operation to the receiving confirmation, in seconds.
If the application is sensitive to latency, such as a database (excessive latency will cause application performance degradation or errors), it is recommended to use SSD storage2. Specific parameter description
[root@host-10-0-1-36 ~]# fio --bs=4k --ioengine=libaio --iodepth=1 --direct=1 --rw=read --time_based --runtime=600 --refill_buffers --norandommap --randrepeat=0 --group_reporting --name=fio-read --size=100G --filename=/dev/vdb
fio-read: (g=0): rw=read, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=1
fio-3.1
Starting 1 process
Jobs: 1 (f=1): [R(1)][100.0%][r=9784KiB/s,w=0KiB/s][r=2446,w=0 IOPS][eta 00m:00s]
fio-read: (groupid=0, jobs=1): err= 0: pid=22004: Wed Oct 10 21:35:42 2018
read: IOPS=2593, BW=10.1MiB/s (10.6MB/s)(6078MiB/600001msec)
slat (usec): min=4, max=1532, avg=11.98, stdev= 8.10
clat (nsec): min=1021, max=66079k, avg=370367.39, stdev=395393.29
lat (usec): min=44, max=66086, avg=382.88, stdev=399.21
clat percentiles (usec):
| 1.00th=[ 42], 5.00th=[ 44], 10.00th=[ 45], 20.00th=[ 46],
| 30.00th=[ 48], 40.00th=[ 51], 50.00th=[ 383], 60.00th=[ 578],
| 70.00th=[ 644], 80.00th=[ 701], 90.00th=[ 783], 95.00th=[ 865],
| 99.00th=[ 988], 99.50th=[ 1020], 99.90th=[ 1336], 99.95th=[ 2057],
| 99.99th=[11338]
bw ( KiB/s): min= 2000, max=68272, per=99.98%, avg=10370.55, stdev=8123.28, samples=1199
iops : min= 500, max=17068, avg=2592.62, stdev=2030.82, samples=1199
lat (usec) : 2=0.01%, 20=0.01%, 50=37.97%, 100=10.63%, 250=0.39%
lat (usec) : 500=6.09%, 750=31.54%, 1000=12.65%
lat (msec) : 2=0.67%, 4=0.03%, 10=0.01%, 20=0.01%, 50=0.01%
lat (msec) : 100=0.01%
cpu : usr=1.43%, sys=5.12%, ctx=1555984, majf=0, minf=35
IO depths : 1=100.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=64=0.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
issued rwt: total=1556003,0,0, short=0,0,0, dropped=0,0,0
latency : target=0, window=0, percentile=100.00%, depth=1
Run status group 0 (all jobs):
READ: bw=10.1MiB/s (10.6MB/s), 10.1MiB/s-10.1MiB/s (10.6MB/s-10.6MB/s), io=6078MiB (6373MB), run=600001-600001msec
Disk stats (read/write):
vdb: ios=1555778/0, merge=0/0, ticks=570518/0, in_queue=569945, util=95.04%| Parameter details | meaning |
|---|---|
| I | How many M IOs were executed |
| bw | Average IO bandwidth |
| iops | IOPS |
| around | Thread running time |
| slat | Submission delay |
| dat | Completion delay |
| years | Response time |
| cpu | cpu utilization |
| IO depths | io queue |
| IO submit | The number of IOs to be submitted for a single IO submission |
| I complete | |
| IO issued | |
| IO latency |
3. FIO test tool parameter description
[root@host-10-0-1-36 ~]# fio -direct=1 -iodepth=64 -rw=read -ioengine=libaio -bs=4096k -size=100G -numjobs=1 -runtime=300 -group_reporting -filename=/dev/vdb -name=Write_PPS_Testing| parameter | Description |
|---|---|
| -direct=1 | Indicates that the I/O cache is ignored during the test, and the data is written directly |
| -iodepth=128 | Indicates that when AIO is used, the upper limit of the number of simultaneous I/Os is 128 |
| -rw = randwrite | Indicates that the read and write strategy during the test is random writes. For other tests, it can be set to: randread (random reads) read (sequential reads) write (sequential writes) randrw (mixed random reads and writes) |
| -ioengine=libaio | Indicates that the test method is libaio (Linux AIO, asynchronous I/O). Applications usually use I/O in two ways: Synchronous: Synchronous I/O can only issue one I/O request at a time, and wait for the kernel to complete before returning. In this way, the iodepth for a single thread is always less than 1, but it can be solved by concurrent execution of multiple threads. Usually 16-32 threads are used to work at the same time to fill the iodepth. Asynchronous: Asynchronous I/O usually uses libaio to submit a batch of I/O requests at a time, and then wait for the completion of the batch to reduce the number of interactions, which will be more efficient. |
| -bs=4k | Indicates that the block file size of a single I/O is 4 KB. The default size when this parameter is not specified is also 4 KB. When testing IOPS, it is recommended to set bs to a relatively small value, such as 4k in this example. When testing throughput, it is recommended to set bs to a larger value, such as 1024k in this example |
| -size=1G | Indicates that the test file size is 1 GiB |
| -numjobs=1 | Indicates that the number of test threads is 1 |
| -runtime=1000 | Indicates that the test time is 1000 seconds. If it is not configured, continue to write the file of the size specified by the aforementioned -size, and write the block size with the -bs value each time |
| -group_reporting | Means that the statistical information of each process is summarized in the test results, rather than the summary display information of different jobs |
| -filename=iotest | Specify the name of the test file, such as iotest. Testing the bare disk can get the real hard disk performance, but testing the bare disk directly will destroy the file system structure, please make a data backup before the test |
| -name=Rand_Write_Testing | Indicates that the test task name is Rand_Write_Testing, which can be set at will |
4. Test results
Here we use 4K data block test random read and random write to test the maximum IOPS, 4M block test sequential read and sequential write to test the maximum throughput
- Test random write of 4K blocks
[root@host-10-0-1-36 ~]# fio --bs=4k --ioengine=libaio --iodepth=128 --direct=1 --rw=randwrite --time_based --runtime=300 --refill_buffers --norandommap --randrepeat=0 --group_reporting --name=fio-write --size=1G --filename=/dev/vdb
fio-write: (g=0): rw=randwrite, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=128
fio-3.1
Starting 1 process
Jobs: 1 (f=1): [w(1)][100.0%][r=0KiB/s,w=3996KiB/s][r=0,w=999 IOPS][eta 00m:00s]
fio-write: (groupid=0, jobs=1): err= 0: pid=22050: Wed Oct 10 22:29:32 2018
write: IOPS=2819, BW=11.0MiB/s (11.5MB/s)(3309MiB/300484msec)
slat (usec): min=2, max=2399, avg= 9.54, stdev=10.19
clat (usec): min=1180, max=3604.0k, avg=45387.25, stdev=168013.09
lat (usec): min=1201, max=3604.0k, avg=45397.35, stdev=168013.57
clat percentiles (usec):
| 1.00th=[ 1713], 5.00th=[ 2212], 10.00th=[ 2835],
| 20.00th=[ 4015], 30.00th=[ 5211], 40.00th=[ 6849],
| 50.00th=[ 8979], 60.00th=[ 11994], 70.00th=[ 17695],
| 80.00th=[ 33162], 90.00th=[ 61604], 95.00th=[ 137364],
| 99.00th=[ 893387], 99.50th=[1266680], 99.90th=[2122318],
| 99.95th=[2432697], 99.99th=[2969568]
bw ( KiB/s): min= 8, max=49120, per=100.00%, avg=11603.11, stdev=10950.79, samples=584
iops : min= 2, max=12280, avg=2900.77, stdev=2737.69, samples=584
lat (msec) : 2=3.16%, 4=16.75%, 10=33.83%, 20=18.49%, 50=14.82%
lat (msec) : 100=6.56%, 250=2.86%, 500=1.49%, 750=0.70%, 1000=0.53%
lat (msec) : 2000=0.69%, >=2000=0.12%
cpu : usr=1.84%, sys=4.16%, ctx=323739, majf=0, minf=28
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.1%
issued rwt: total=0,847133,0, short=0,0,0, dropped=0,0,0
latency : target=0, window=0, percentile=100.00%, depth=128
Run status group 0 (all jobs):
WRITE: bw=11.0MiB/s (11.5MB/s), 11.0MiB/s-11.0MiB/s (11.5MB/s-11.5MB/s), io=3309MiB (3470MB), run=300484-300484msec
Disk stats (read/write):
vdb: ios=91/847074, merge=0/0, ticks=3/38321566, in_queue=38360706, util=100.00%
我们可以看到4K的数据块，随机写的最大IOPS为：12280
- 测试4K的随机读
[root@host-10-0-1-36 ~]# fio --bs=4k --ioengine=libaio --iodepth=128 --direct=1 --rw=randread --time_based --runtime=300 --refill_buffers --norandommap --randrepeat=0 --group_reporting --name=fio-write --size=1G --filename=/dev/vdb
fio-write: (g=0): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=128
fio-3.1
Starting 1 process
Jobs: 1 (f=1): [r(1)][100.0%][r=87.1MiB/s,w=0KiB/s][r=22.3k,w=0 IOPS][eta 00m:00s]
fio-write: (groupid=0, jobs=1): err= 0: pid=22055: Wed Oct 10 22:51:54 2018
read: IOPS=22.3k, BW=87.0MiB/s (91.2MB/s)(25.5GiB/300004msec)
slat (usec): min=2, max=8626, avg= 7.91, stdev=11.85
clat (usec): min=36, max=71810, avg=5735.17, stdev=1405.20
lat (usec): min=45, max=71826, avg=5743.59, stdev=1405.30
clat percentiles (usec):
| 1.00th=[ 1958], 5.00th=[ 3556], 10.00th=[ 4424], 20.00th=[ 5145],
| 30.00th=[ 5407], 40.00th=[ 5604], 50.00th=[ 5735], 60.00th=[ 5866],
| 70.00th=[ 6063], 80.00th=[ 6259], 90.00th=[ 6783], 95.00th=[ 7504],
| 99.00th=[10290], 99.50th=[12256], 99.90th=[16712], 99.95th=[18482],
| 99.99th=[25035]
bw ( KiB/s): min=74872, max=93240, per=100.00%, avg=89121.48, stdev=2687.55, samples=600
iops : min=18718, max=23310, avg=22280.35, stdev=671.89, samples=600
lat (usec) : 50=0.01%, 100=0.01%, 250=0.01%, 500=0.02%, 750=0.06%
lat (usec) : 1000=0.12%
lat (msec) : 2=0.86%, 4=6.02%, 10=91.77%, 20=1.11%, 50=0.03%
lat (msec) : 100=0.01%
cpu : usr=6.56%, sys=28.57%, ctx=3473136, majf=0, minf=160
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.1%
issued rwt: total=6683408,0,0, short=0,0,0, dropped=0,0,0
latency : target=0, window=0, percentile=100.00%, depth=128
Run status group 0 (all jobs):
READ: bw=87.0MiB/s (91.2MB/s), 87.0MiB/s-87.0MiB/s (91.2MB/s-91.2MB/s), io=25.5GiB (27.4GB), run=300004-300004msec
Disk stats (read/write):
vdb: ios=6680955/0, merge=0/0, ticks=37981396/0, in_queue=37983491, util=100.00%我们可以看到4K的数据块，随机读的最大IOPS为：23310
- 测试4M的顺序写
[root@host-10-0-1-36 ~]# fio -direct=1 -iodepth=64 -rw=write -ioengine=libaio -bs=4096k -size=100G -numjobs=1 -runtime=300 -group_reporting -filename=/dev/vdb -name=Write_PPS_Testing
Write_PPS_Testing: (g=0): rw=write, bs=(R) 4096KiB-4096KiB, (W) 4096KiB-4096KiB, (T) 4096KiB-4096KiB, ioengine=libaio, iodepth=64
fio-3.1
Starting 1 process
Jobs: 1 (f=1): [W(1)][100.0%][r=0KiB/s,w=108MiB/s][r=0,w=27 IOPS][eta 00m:00s]
Write_PPS_Testing: (groupid=0, jobs=1): err= 0: pid=22098: Wed Oct 10 23:20:07 2018
write: IOPS=28, BW=115MiB/s (120MB/s)(33.7GiB/300466msec)
slat (usec): min=564, max=2554.1k, avg=34767.87, stdev=65067.32
clat (msec): min=400, max=12179, avg=2188.04, stdev=1037.46
lat (msec): min=473, max=12231, avg=2222.81, stdev=1047.45
clat percentiles (msec):
| 1.00th=[ 1435], 5.00th=[ 1586], 10.00th=[ 1653], 20.00th=[ 1754],
| 30.00th=[ 1838], 40.00th=[ 1921], 50.00th=[ 2005], 60.00th=[ 2089],
| 70.00th=[ 2198], 80.00th=[ 2333], 90.00th=[ 2534], 95.00th=[ 2802],
| 99.00th=[ 8490], 99.50th=[ 9731], 99.90th=[11745], 99.95th=[12013],
| 99.99th=[12147]
bw ( KiB/s): min= 8192, max=196608, per=100.00%, avg=120954.04, stdev=31456.39, samples=580
iops : min= 2, max= 48, avg=29.53, stdev= 7.68, samples=580
lat (msec) : 500=0.02%, 750=0.03%, 1000=0.13%, 2000=49.25%, >=2000=50.56%
cpu : usr=1.04%, sys=2.16%, ctx=6387, majf=0, minf=29
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.2%, 32=0.4%, >=64=99.3%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
issued rwt: total=0,8627,0, short=0,0,0, dropped=0,0,0
latency : target=0, window=0, percentile=100.00%, depth=64
Run status group 0 (all jobs):
WRITE: bw=115MiB/s (120MB/s), 115MiB/s-115MiB/s (120MB/s-120MB/s), io=33.7GiB (36.2GB), run=300466-300466msec
Disk stats (read/write):
vdb: ios=91/77596, merge=0/0, ticks=6/37685137, in_queue=37717833, util=99.99%我们可以看到4M的数据块，顺序写的最大吞吐量为：196M
- 测试4M的顺序读
[root@host-10-0-1-36 ~]# fio -direct=1 -iodepth=64 -rw=read -ioengine=libaio -bs=4096k -size=100G -numjobs=1 -runtime=300 -group_reporting -filename=/dev/vdb -name=Write_PPS_Testing
Write_PPS_Testing: (g=0): rw=read, bs=(R) 4096KiB-4096KiB, (W) 4096KiB-4096KiB, (T) 4096KiB-4096KiB, ioengine=libaio, iodepth=64
fio-3.1
Starting 1 process
Jobs: 1 (f=1): [R(1)][100.0%][r=312MiB/s,w=0KiB/s][r=78,w=0 IOPS][eta 00m:00s]
Write_PPS_Testing: (groupid=0, jobs=1): err= 0: pid=22103: Wed Oct 10 23:26:31 2018
read: IOPS=35, BW=142MiB/s (149MB/s)(41.6GiB/300201msec)
slat (usec): min=469, max=95190, avg=28166.40, stdev=15966.53
clat (msec): min=185, max=4070, avg=1772.98, stdev=551.21
lat (msec): min=205, max=4107, avg=1801.15, stdev=558.25
clat percentiles (msec):
| 1.00th=[ 518], 5.00th=[ 634], 10.00th=[ 751], 20.00th=[ 1536],
| 30.00th=[ 1770], 40.00th=[ 1854], 50.00th=[ 1921], 60.00th=[ 1989],
| 70.00th=[ 2039], 80.00th=[ 2140], 90.00th=[ 2299], 95.00th=[ 2433],
| 99.00th=[ 2802], 99.50th=[ 2970], 99.90th=[ 3473], 99.95th=[ 3641],
| 99.99th=[ 3775]
bw ( KiB/s): min=106496, max=466944, per=99.97%, avg=145216.28, stdev=60136.24, samples=597
iops : min= 26, max= 114, avg=35.45, stdev=14.68, samples=597
lat (msec) : 250=0.10%, 500=0.66%, 750=9.21%, 1000=7.02%, 2000=46.51%
lat (msec) : >=2000=36.51%
cpu : usr=0.05%, sys=2.61%, ctx=10959, majf=0, minf=672
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.2%, 32=0.3%, >=64=99.4%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.1%, >=64=0.0%
issued rwt: total=10646,0,0, short=0,0,0, dropped=0,0,0
latency : target=0, window=0, percentile=100.00%, depth=64
Run status group 0 (all jobs):
READ: bw=142MiB/s (149MB/s), 142MiB/s-142MiB/s (149MB/s-149MB/s), io=41.6GiB (44.7GB), run=300201-300201msec
Disk stats (read/write):
vdb: ios=95740/0, merge=0/0, ticks=37360788/0, in_queue=37382226, util=100.00%我们可以看到4M的数据块，顺序读的最大吞吐量为：466M