为什么在我将节点添加到 Hadoop 集群时我的吞吐量和平均 io 速率变慢了？

Question

所以我运行在我的集群上使用 TestDFSIO 来查看读写操作的吞吐量和平均 io 速率。我做了 4 个测试： 4 个文件，每个 256 MB（总计 1 GB） 2 个文件，每个 256 MB（总计 512 MB） 2 个文件，每个 128 MB（总计 256 MB） 1 个文件 50 MB（总计 50 MB）

和我运行那些在单节点到 5 节点 hadoop 集群上。块大小为 256 MB，每个节点都有不同的复制（单节点 = 1 个复制，2 个节点 = 2 个复制，依此类推）。

这里是1GB数据测试的测试结果 1 个节点

----- TestDFSIO ----- : write
           Date & time: Thu Jun 20 11:38:21 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 8.503288381053611
Average IO rate mb/sec: 8.507380485534668
 IO rate std deviation: 0.18595730311606032
    Test exec time sec: 84.876

----- TestDFSIO ----- : read
           Date & time: Thu Jun 20 11:39:52 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 14.351786965662228
Average IO rate mb/sec: 14.422638893127441
 IO rate std deviation: 1.0515649052955383
    Test exec time sec: 61.371

2 node
----- TestDFSIO ----- : write
           Date & time: Thu Jun 20 11:15:52 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 2.557167936510315
Average IO rate mb/sec: 2.5574562549591064
 IO rate std deviation: 0.027311795003682558
    Test exec time sec: 150.506

----- TestDFSIO ----- : read
           Date & time: Thu Jun 20 11:18:04 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 9.567321617101587
Average IO rate mb/sec: 9.673456192016602
 IO rate std deviation: 1.0593562755825534
    Test exec time sec: 79.333

3 node
----- TestDFSIO ----- : write
           Date & time: Thu Jun 20 10:42:47 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 2.343067129788529
Average IO rate mb/sec: 2.3866918087005615
 IO rate std deviation: 0.3233444726530288
    Test exec time sec: 167.593

----- TestDFSIO ----- : read
           Date & time: Thu Jun 20 10:47:33 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 11.901164547546546
Average IO rate mb/sec: 12.255699157714844
 IO rate std deviation: 2.2415787547598667
    Test exec time sec: 69.29

4 node 
----- TestDFSIO ----- : write
           Date & time: Thu Jun 20 10:23:19 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 1.6539390885245053
Average IO rate mb/sec: 1.6625666618347168
 IO rate std deviation: 0.12093049037575003
    Test exec time sec: 205.164

----- TestDFSIO ----- : read
           Date & time: Thu Jun 20 10:25:23 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 19.842653954966476
Average IO rate mb/sec: 20.02923583984375
 IO rate std deviation: 1.9719328195872965
    Test exec time sec: 57.25

5 node
----- TestDFSIO ----- : write
           Date & time: Thu Jun 13 12:50:12 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 1.5617159964556366
Average IO rate mb/sec: 1.573684573173523
 IO rate std deviation: 0.14426118715726127
    Test exec time sec: 219.959

----- TestDFSIO ----- : read
           Date & time: Thu Jun 13 14:01:01 WIB 2019
       Number of files: 4
Total MBytes processed: 1024.0
     Throughput mb/sec: 18.00692844707827
Average IO rate mb/sec: 18.323461532592773
 IO rate std deviation: 2.501963465819598
    Test exec time sec: 64.316

我认为节点越多，工作就越并行化并增加了吞吐量。为什么添加新节点时写入操作会大幅下降？

Answer 1

您的数据量太小。单个系统可以轻松处理 1 GB 的数据。考虑到这是您使用的最大尺寸，您看到这些结果也就不足为奇了。

将这个数量级提高到 100GB-1TB 之类的东西，否则从这种类型的测试中得出性能结果没有任何意义。

为什么在我将节点添加到 Hadoop 集群时我的吞吐量和平均 io 速率变慢了？

Why is my throughput and average io rate got slower when i add node to my Hadoop cluster?

hadoop

io

mapreduce

benchmarking

throughput