[转帖]如何对minio进行性能测试和分析

https://developer.aliyun.com/article/1006775

环境详情

server(组成集群，ec为12:4)

ip	hosts	硬盘
storage01	172.16.50.1	12*10T
storage02	172.16.50.2	12*10T
storage03	172.16.50.3	12*10T
storage04	172.16.50.4	12*10T

client

ip	host
headnode	172.16.50.5
node02	172.16.50.6
node03	172.16.50.7
node04	172.16.50.8

为什么选择speedtest和warp作为测试工具？

speedtest是一个易用的测试工具，它会先运行PUTS，然后运行GETS，通过增量的方式测试得到最大吞吐量。而warp则是一个完整的工具链，提供了很独立的测试项，能够测试GET;PUT;DELETE等都可以测试得到。同时通过cs的结构设计，更符合真实的使用场景，得到最贴近应用的性能结果，有利于性能分析。

warp结构如下图所示

环境详情

server(组成集群，ec为12:4)

ip	hosts	硬盘
storage01	172.16.50.1	12*10T
storage02	172.16.50.2	12*10T
storage03	172.16.50.3	12*10T
storage04	172.16.50.4	12*10T

client

ip	host
headnode	172.16.50.5
node02	172.16.50.6
node03	172.16.50.7
node04	172.16.50.8

为什么选择speedtest和warp作为测试工具？

warp结构如下图所示

开始测试

speedtest

开始前需要下载minio client

wget https://dl.min.io/client/mc/release/linux-amd64/mc

chmod +x mc

mv mc /usr/local/bin/

写入host到/etc/hosts
配置客户端

/usr/local/bin/mc alias set minio http://172.16.50.1:9000 <YOUR-ACCESS-KEY> <YOUR-SECRET-KEY>

开始测试，执行（以前叫speedtest，现在修改成support perf了）

/usr/local/bin/mc support perf minio/<bucket>

这里可以得到一个网络、硬盘和吞吐量的结果

NODE                    RX              TX

http://storage01:9000   1.8 GiB/s       2.0 GiB/s

http://storage02:9000   1.8 GiB/s       2.0 GiB/s

http://storage03:9000   2.7 GiB/s       2.5 GiB/s

http://storage04:9000   2.4 GiB/s       2.2 GiB/s

NetPerf: 

NODE                    PATH    READ            WRITE

http://storage02:9000   /data1  192 MiB/s       235 MiB/s

http://storage02:9000   /data2  219 MiB/s       240 MiB/s

http://storage02:9000   /data3  186 MiB/s       262 MiB/s

http://storage02:9000   /data4  191 MiB/s       230 MiB/s

http://storage02:9000   /data5  179 MiB/s       221 MiB/s

http://storage02:9000   /data6  170 MiB/s       222 MiB/s

http://storage02:9000   /data7  200 MiB/s       219 MiB/s

http://storage02:9000   /data8  198 MiB/s       230 MiB/s

http://storage02:9000   /data9  166 MiB/s       207 MiB/s

http://storage02:9000   /data10 206 MiB/s       209 MiB/s

http://storage02:9000   /data11 198 MiB/s       213 MiB/s

http://storage02:9000   /data12 164 MiB/s       205 MiB/s

http://storage04:9000   /data1  172 MiB/s       259 MiB/s

http://storage04:9000   /data2  201 MiB/s       250 MiB/s

http://storage04:9000   /data3  218 MiB/s       256 MiB/s

http://storage04:9000   /data4  188 MiB/s       245 MiB/s

http://storage04:9000   /data5  157 MiB/s       197 MiB/s

http://storage04:9000   /data6  155 MiB/s       205 MiB/s

http://storage04:9000   /data7  154 MiB/s       210 MiB/s

http://storage04:9000   /data8  143 MiB/s       185 MiB/s

http://storage04:9000   /data9  181 MiB/s       207 MiB/s

http://storage04:9000   /data10 174 MiB/s       214 MiB/s

http://storage04:9000   /data11 173 MiB/s       218 MiB/s

http://storage04:9000   /data12 178 MiB/s       206 MiB/s

http://storage03:9000   /data1  194 MiB/s       337 MiB/s

http://storage03:9000   /data2  204 MiB/s       267 MiB/s

http://storage03:9000   /data3  212 MiB/s       261 MiB/s

http://storage03:9000   /data4  200 MiB/s       235 MiB/s

http://storage03:9000   /data5  168 MiB/s       216 MiB/s

http://storage03:9000   /data6  209 MiB/s       221 MiB/s

http://storage03:9000   /data7  179 MiB/s       222 MiB/s

http://storage03:9000   /data8  170 MiB/s       220 MiB/s

http://storage03:9000   /data9  144 MiB/s       186 MiB/s

http://storage03:9000   /data10 142 MiB/s       172 MiB/s

http://storage03:9000   /data11 149 MiB/s       171 MiB/s

http://storage03:9000   /data12 132 MiB/s       206 MiB/s

http://storage01:9000   /data1  99 MiB/s        119 MiB/s

http://storage01:9000   /data2  103 MiB/s       114 MiB/s

http://storage01:9000   /data3  104 MiB/s       114 MiB/s

http://storage01:9000   /data4  98 MiB/s        186 MiB/s

http://storage01:9000   /data5  186 MiB/s       261 MiB/s

http://storage01:9000   /data6  224 MiB/s       251 MiB/s

http://storage01:9000   /data7  188 MiB/s       240 MiB/s

http://storage01:9000   /data8  222 MiB/s       252 MiB/s

http://storage01:9000   /data9  185 MiB/s       247 MiB/s

http://storage01:9000   /data10 209 MiB/s       236 MiB/s

http://storage01:9000   /data11 191 MiB/s       224 MiB/s

http://storage01:9000   /data12 222 MiB/s       250 MiB/s

DrivePerf: 

        THROUGHPUT      IOPS

PUT     1.3 GiB/s       21 objs/s

GET     4.2 GiB/s       67 objs/s

MinIO 2022-08-25T07:17:05Z, 4 servers, 48 drives, 64 MiB objects, 62 threads

ObjectPerf:

我们其实还可以按需做一些调整，做一些独立的测试。比如只需要测量object的读写速度，可以使用下述命令

mc support perf object minio

同理，我们也可以单独测量硬盘或者网络吞吐量，如下述命令

# driver

root@storage01:~# mc support perf object minio

        THROUGHPUT      IOPS

PUT     1.6 GiB/s       26 objs/s

GET     4.0 GiB/s       63 objs/s

MinIO 2022-08-25T07:17:05Z, 4 servers, 48 drives, 64 MiB objects, 27 threads

ObjectPerf:

# throughtput

root@storage01:~# mc support perf net minio

NODE                    RX              TX

http://storage01:9000   1.9 GiB/s       2.1 GiB/s

http://storage02:9000   2.4 GiB/s       2.2 GiB/s

http://storage03:9000   2.6 GiB/s       2.4 GiB/s

http://storage04:9000   2.1 GiB/s       2.3 GiB/s

NetPerf:

也可以选择在指定时间内测试指定大小的对象的读写速度

root@storage01:~# mc support perf object minio --duration 20s --size 128MiB

        THROUGHPUT      IOPS

PUT     2.3 GiB/s       18 objs/s

GET     3.8 GiB/s       30 objs/s

MinIO 2022-08-25T07:17:05Z, 4 servers, 48 drives, 128 MiB objects, 41 threads

ObjectPerf:

warp

在开始测试之前，需要在客户端上安装warp，GitHub - minio/warp: S3 benchmarking tool这里面有多个版本可供选择，建议使用二进制包，比较省事。

wget https://github.com/minio/warp/releases/download/v0.6.2/warp_0.6.2_Linux_x86_64.tar.gz

mkdir warp

tar zxf warp_0.6.2_Linux_x86_64.tar.gz -C warp

# 执行客户端

./warp client

执行后，默认开启7761端口，假如配置了防火墙还需要放行该端口才行，我这里没有配置，所以就不演示了

在所有客户端开启warp后，我们可以选择启动一台执行warp混合基准测试了

warp mixed --duration=3m --warp-client=headnode --warp-client=node0{2...3} --host=storage0{1...4}:9000 --access-key=minio --secret-key=miniodev

warp: Benchmark data written to "warp-remote-2022-09-03[172425]-Fj1O.csv.zst"

Mixed operations.

Operation: DELETE, 10%, Concurrency: 40, Ran 2m58s.

 * Throughput: 18.47 obj/s

Operation: GET, 45%, Concurrency: 40, Ran 2m58s.

 * Throughput: 831.05 MiB/s, 83.11 obj/s

Operation: PUT, 15%, Concurrency: 40, Ran 2m58s.

 * Throughput: 276.45 MiB/s, 27.65 obj/s

Operation: STAT, 30%, Concurrency: 40, Ran 2m58s.

 * Throughput: 55.30 obj/s

Cluster Total: 1106.95 MiB/s, 184.46 obj/s over 2m58s.

我们也可以单独对GET操作进行压力测试得到最大的写入吞吐量

接下来我们对DELETE操作进行测试

root@headnode:~# warp delete --duration=3m --warp-client=headnode --warp-client=node0{2...3} --host=storage0{1...4}:9000 --access-key=minio --secret-key=miniodev

warp: Benchmark data written to "warp-remote-2022-09-03[173528]-1d62.csv.zst"

----------------------------------------

Operation: PUT

* Average: 0.56 MiB/s, 576.46 obj/s

Throughput by host:

 * http://storage01:9000: Avg: 0.14 MiB/s, 144.36 obj/s

 * http://storage02:9000: Avg: 0.14 MiB/s, 143.90 obj/s

 * http://storage03:9000: Avg: 0.14 MiB/s, 143.78 obj/s

 * http://storage04:9000: Avg: 0.14 MiB/s, 144.32 obj/s

Throughput, split into 86 x 1s:

 * Fastest: 655.8KiB/s, 655.89 obj/s

 * 50% Median: 584.4KiB/s, 584.43 obj/s

 * Slowest: 438.5KiB/s, 438.49 obj/s

----------------------------------------

Operation: DELETE

* Average: 897.41 obj/s

Throughput by host:

 * http://storage01:9000: Avg: 225.39 obj/s

 * http://storage02:9000: Avg: 221.08 obj/s

 * http://storage03:9000: Avg: 220.70 obj/s

 * http://storage04:9000: Avg: 222.31 obj/s

Throughput, split into 41 x 1s:

 * Fastest: 998.26 obj/s

 * 50% Median: 890.61 obj/s

 * Slowest: 861.53 obj/s

warp: Cleanup done

对DELETE操作进行测试

root@headnode:~# warp stat --autoterm --duration=3m --warp-client=headnode --warp-client=node0{2...3} --host=storage0{1...4}:9000 --access-key=minio --secret-key=miniodev

warp: Benchmark data written to "warp-remote-2022-09-03[173008]-vsV9.csv.zst"

----------------------------------------

Operation: PUT

* Average: 0.56 MiB/s, 584.19 obj/s

Throughput by host:

 * http://storage01:9000: Avg: 0.14 MiB/s, 145.46 obj/s

 * http://storage02:9000: Avg: 0.14 MiB/s, 145.46 obj/s

 * http://storage03:9000: Avg: 0.14 MiB/s, 145.32 obj/s

 * http://storage04:9000: Avg: 0.14 MiB/s, 144.19 obj/s

Throughput, split into 33 x 1s:

 * Fastest: 609.4KiB/s, 624.07 obj/s

 * 50% Median: 585.7KiB/s, 599.81 obj/s

 * Slowest: 499.3KiB/s, 511.37 obj/s

----------------------------------------

Operation: STAT

* Average: 10262.16 obj/s

Throughput by host:

 * http://storage01:9000: Avg: 2549.94 obj/s

 * http://storage02:9000: Avg: 2566.62 obj/s

 * http://storage03:9000: Avg: 2572.47 obj/s

 * http://storage04:9000: Avg: 2572.58 obj/s

Throughput, split into 179 x 1s:

 * Fastest: 10622.95 obj/s

 * 50% Median: 10268.04 obj/s

 * Slowest: 9574.31 obj/s

warp: Cleanup done.

将文件切片分批上传性能测试

root@headnode:~# warp multipart --parts=500 --part.size=10MiB --warp-client=headnode --warp-client=node0{2...3} --host=storage0{1...4}:9000 --access-key=minio --secret-key=miniodev

warp: Benchmark data written to "warp-remote-2022-09-03[174331]-AWo5.csv.zst"

----------------------------------------

Operation: PUT

* Average: 559.26 MiB/s, 55.93 obj/s

Throughput by host:

 * http://storage01:9000: Avg: 137.98 MiB/s, 13.80 obj/s

 * http://storage02:9000: Avg: 140.59 MiB/s, 14.06 obj/s

 * http://storage03:9000: Avg: 138.60 MiB/s, 13.86 obj/s

 * http://storage04:9000: Avg: 141.93 MiB/s, 14.19 obj/s

Throughput, split into 16 x 1s:

 * Fastest: 616.2MiB/s, 61.62 obj/s

 * 50% Median: 562.9MiB/s, 56.29 obj/s

 * Slowest: 437.3MiB/s, 43.73 obj/s

----------------------------------------

Operation: GET

* Average: 1026.15 MiB/s, 102.62 obj/s

Throughput by host:

 * http://storage01:9000: Avg: 254.73 MiB/s, 25.47 obj/s

 * http://storage02:9000: Avg: 257.86 MiB/s, 25.79 obj/s

 * http://storage03:9000: Avg: 257.91 MiB/s, 25.79 obj/s

 * http://storage04:9000: Avg: 255.68 MiB/s, 25.57 obj/s

Throughput, split into 298 x 1s:

 * Fastest: 1133.4MiB/s, 113.34 obj/s

 * 50% Median: 1030.0MiB/s, 103.00 obj/s

 * Slowest: 876.0MiB/s, 87.60 obj/s

warp: Cleanup done.

好的，以上就是我做的一些测试和总结，希望对大家有帮助