hadoop性能测试

1、hadoop自带的性能基准评测工具 

（一）TestDFSIO 一、测试写性能 （1）如有必要，先删除历史数据 $hadoop jar /home/hadoop/hadoop/share/hadoop/mapreduce2/hadoop-mapreduce-client-jobclient-2.3.0-cdh5.1.2-tests.jar TestDFSIO -clean （2）执行测试 $hadoop jar /home/hadoop/hadoop/share/hadoop/mapreduce2/hadoop-mapreduce-client-jobclient-2.3.0-cdh5.1.2-tests.jar TestDFSIO -write -nrFiles 5 -fileSize 20 （3）查看结果：每一次测试生成一个结果，并以附加的形式添加到TestDFSIO_results.log中 $cat TestDFSIO_results.log ----- TestDFSIO ----- : write            Date & time: Mon May 11 09:41:34 HKT 2015        Number of files: Total MBytes processed: 100.0      Throughput mb/sec: 21.468441391155004 Average IO rate mb/sec: 25.366744995117188  IO rate std deviation: 12.744636924030177     Test exec time sec: 27.585 ----- TestDFSIO ----- : write            Date & time: Mon May 11 09:42:28 HKT 2015        Number of files: 5 Total MBytes processed: 100.0      Throughput mb/sec: 22.779043280182233 Average IO rate mb/sec: 25.440486907958984  IO rate std deviation: 9.930490103638768     Test exec time sec: 26.67 （4）结果说明 Total MBytes processed ： 总共须要写入的数据量 100MB Throughput mb/sec ：总共须要写入的数据量/（每一个map任务实际写入数据的执行时间之和（这个时间会远小于Test exec time sec））＝＝》100/(map1写时间+map2写时间+...) Average IO rate mb/sec ：（每一个map须要写入的数据量/每一个map任务实际写入数据的执行时间）之和/任务数＝＝》(20/map1写时间＋20/map2写时间+...)/1000，因此这个值跟上面一个值老是存在差别。 IO rate std deviation ：上一个值的标准差 Test exec time sec ：整个job的执行时间 二、测试读性能 （1）执行测试 $ hadoop jar /home/hadoop/hadoop/share/hadoop/mapreduce2/hadoop-mapreduce-client-jobclient-2.3.0-cdh5.1.2-tests.jar TestDFSIO -read -nrFiles 5 -fileSize 20 （2）查看测试结果 $ cat TestDFSIO_results.log ----- TestDFSIO ----- : read            Date & time: Mon May 11 09:53:27 HKT 2015        Number of files: 5 Total MBytes processed: 100.0      Throughput mb/sec: 534.75935828877 Average IO rate mb/sec: 540.4888916015625  IO rate std deviation: 53.93029580221512     Test exec time sec: 26.704 （3）结果说明 结果各项意思与write相同，但其读速率比写速率快不少，而总执行时间很是接近。真正测试时，应该用较大的数据量来执行，才可体现出两者的差别。 （二）排序测试 在api文档中搜索terasort，可查询相关信息。 排序测试的三个基本步骤： 生成随机数据??>排序??>验证排序结果 关于terasort更详细的原理，见http://blog.csdn.net/yuesichiu/article/details/17298563 一、生成随机数据 $ hadoop jar /home/hadoop/hadoop/share/hadoop/mapreduce2/hadoop-mapreduce-examples-2.3.0-cdh5.1.2.jar  teragen -Dmapreduce.job.maps=5 10000000 /tmp/hadoop/terasort 此步骤将在hdfs中的 /tmp/hadoop/terasort  中生成数据， $  hadoop fs -ls /tmp/hadoop/terasort Found 6 items -rw-r-----   3 hadoop supergroup          0 2015-05-11 11:32 /tmp/hadoop/terasort/_SUCCESS -rw-r-----   3 hadoop supergroup  200000000 2015-05-11 11:32 /tmp/hadoop/terasort/part-m-00000 -rw-r-----   3 hadoop supergroup  200000000 2015-05-11 11:32 /tmp/hadoop/terasort/part-m-00001 -rw-r-----   3 hadoop supergroup  200000000 2015-05-11 11:32 /tmp/hadoop/terasort/part-m-00002 -rw-r-----   3 hadoop supergroup  200000000 2015-05-11 11:32 /tmp/hadoop/terasort/part-m-00003 -rw-r-----   3 hadoop supergroup  200000000 2015-05-11 11:32 /tmp/hadoop/terasort/part-m-00004 $ hadoop fs -du -s -h /tmp/hadoop/terasort 953.7 M  /tmp/hadoop/terasort 生成的5个数据居然是每一个200M，未解，为何不是10M？？？ 二、运行测试 $hadoop jar /home/hadoop/hadoop/share/hadoop/mapreduce2/hadoop-mapreduce-examples-2.3.0-cdh5.1.2.jar  terasort -Dmapreduce.job.maps=5 /tmp/hadoop/terasort /tmp/hadoop/terasort_out Spent 354ms computing base-splits. Spent 8ms computing TeraScheduler splits. Computing input splits took 365ms Sampling 10 splits of 10 Making 1 from 100000 sampled records Computing parititions took 6659ms Spent 7034ms computing partitions. 三、验证结果  $ hadoop jar /home/hadoop/hadoop/share/hadoop/mapreduce2/hadoop-mapreduce-examples-2.3.0-cdh5.1.2.jar teravalidate  /tmp/hadoop/terasort_out /tmp/hadoop/terasort_report Spent 44ms computing base-splits. Spent 7ms computing TeraScheduler splits. 2、hibench hibench4.0测试不成功，使用3.0代替 一、下载并解压 wget https://codeload.github.com/intel-hadoop/HiBench/zip/HiBench-3.0.0 unzip HiBench-3.0.0 二、修改文件  bin/hibench-config.sh，主要是这几个 export JAVA_HOME=/home/hadoop/jdk1.7.0_67 export HADOOP_HOME=/home/hadoop/hadoop export HADOOP_EXECUTABLE=/home/hadoop/hadoop//bin/hadoop export HADOOP_CONF_DIR=/home/hadoop/conf export HADOOP_EXAMPLES_JAR=/home/hadoop/hadoop/share/hadoop/mapreduce2/hadoop-mapreduce-examples-2.3.0-cdh5.1.2.jar export MAPRED_EXECUTABLE=/home/hadoop/hadoop/bin/mapred #Set the varaible below only in YARN mode export HADOOP_JOBCLIENT_TESTS_JAR=/home/hadoop/hadoop/share/hadoop/mapreduce2/hadoop-mapreduce-examples-2.3.0-cdh5.1.2.jar/hadoop-mapreduce-client-jobclient-2.3.0-cdh5.1.2-tests.jar 三、修改conf/benchmarks.lst，哪些不想运行的将之注释掉 四、运行 bin/run-all.sh 五、查看结果 在当前目录会生成hibench.report文件，内容以下 Type         Date       Time     Input_data_size      Duration(s)          Throughput(bytes/s)  Throughput/node WORDCOUNT    2015-05-12 19:32:33 251.248 DFSIOE-READ  2015-05-12 19:54:29 54004092852          463.863              116422505            38807501 DFSIOE-WRITE 2015-05-12 20:02:57 27320849148          498.132              54846605             18282201 PAGERANK     2015-05-12 20:27:25 711.391 SORT         2015-05-12 20:33:21 243.603 TERASORT     2015-05-12 20:40:34 10000000000          266.796              37481821             12493940 SLEEP        2015-05-12 20:40:40 0                    .177                 0                    0