hadoop的安装配置这里就不讲了。
Sqoop的安装也很简单。 完成sqoop的安装后,可以这样测试是否可以连接到mysql(注意:mysql的jar包要放到 SQOOP_HOME/lib 下): sqoop list-databases --connect jdbc:mysql://192.168.1.109:3306/ --username root --password 19891231 结果如下

即说明sqoop已经可以正常使用了。 下面,要将mysql中的数据导入到hadoop中。 我准备的是一个300万条数据的身份证数据表:

先启动hive(使用命令行:hive 即可启动) 然后使用sqoop导入数据到hive: sqoop import --connect jdbc:mysql://192.168.1.109:3306/hadoop --username root --password 19891231 --table test_sfz --hive-import sqoop 会启动job来完成导入工作。

完成导入用了2分20秒,还是不错的。 在hive中可以看到刚刚导入的数据表:

我们来一句sql测试一下数据: select * from test_sfz where id < 10;

可以看到,hive完成这个任务用了将近25秒,确实是挺慢的(在mysql中几乎是不费时间),但是要考虑到hive是创建了job在hadoop中跑,时间当然多。
接下来,我们会对这些数据进行复杂查询的测试: 我机子的配置如下:

hadoop 是运行在虚拟机上的伪分布式,虚拟机OS是ubuntu12.04 64位,配置如下:
TEST 1 计算平均年龄
测试数据:300.8 W 1. 计算广东的平均年龄 mysql:select (sum(year(NOW()) - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz where address like '广东%'; 用时: 0.877s hive:select (sum(year('2014-10-01') - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz where address like '广东%'; 用时:25.012s 2. 对每个城市的的平均年龄进行从高到低的排序 mysql:select address, (sum(year(NOW()) - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz GROUP BY address order by ageAvge desc; 用时:2.949s hive:select address, (sum(year('2014-10-01') - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz GROUP BY address order by ageAvge desc; 用时:51.29s 可以看到,在耗时上面,hive的增长速度较mysql慢。
TEST 2
测试数据:1200W mysql 引擎: MyISAM(为了加快查询速度) 导入到hive:

1. 计算广东的平均年龄 mysql:select (sum(year(NOW()) - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz2 where address like '广东%'; 用时: 5.642s hive:select (sum(year('2014-10-01') - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz2 where address like '广东%'; 用时:168.259s 2. 对每个城市的的平均年龄进行从高到低的排序 mysql:select address, (sum(year(NOW()) - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz2 GROUP BY address order by ageAvge desc; 用时:11.964s hive:select address, (sum(year('2014-10-01') - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz2 GROUP BY address order by ageAvge desc; 用时:311.714s
TEST 3
测试数据:2000W mysql 引擎: MyISAM(为了加快查询速度) 导入到hive:

(这次用的时间很短!可能是因为TEST2中的导入时,我的主机在做其他耗资源的工作..) 1. 计算广东的平均年龄 mysql:select (sum(year(NOW()) - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz2 where address like '广东%'; 用时: 6.605s hive:select (sum(year('2014-10-01') - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz2 where address like '广东%'; 用时:188.206s 2. 对每个城市的的平均年龄进行从高到低的排序 mysql:select address, (sum(year(NOW()) - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz2 GROUP BY address order by ageAvge desc; 用时:19.926s hive:select address, (sum(year('2014-10-01') - SUBSTRING(borth,1,4))/count(*)) as ageAvge from test_sfz2 GROUP BY address order by ageAvge desc; 用时:411.816s