关于关系表的设计归根结底有两个方面。

第一，就是完全按照范式理论去设计，一般来说达到第三范式就可以了，或者你可以划分的更细到达更上一层次。比如第四，第五，第六等等。这种设计有自己的可读性很强，但是有一点，在检索数据的时候增加了多张关系表来做关联的开销。

第二，就是在范式理论上适当的做些反范式，有的东西还是不要太剥离的好。（窄表以及宽表） 这点和软件设计中的紧耦合松耦合理论一致。

下面我就以常用的LOG表来做下演示，其中有两种表的实际，一种是窄表，一种是稍微宽一点的表。

窄表：log_ytt

mysql> show create table log_ytt;    +-------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------+| Table       | Create Table                                                                                                                                                             |+-------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------+| log_ytt | CREATE TABLE `log_ytt` (  `ids` bigint(20) DEFAULT NULL,  `log_time` datetime DEFAULT NULL,  KEY `idx_u1` (`ids`,`log_time`)) ENGINE=InnoDB DEFAULT CHARSET=utf8 | +-------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------+1 row in set (0.00 sec)

表记录数

mysql>  select * from log_ytt where ids > '4875000001';                                                                                    +------------+---------------------+| ids        | log_time            |+------------+---------------------+| 7110000001 | 2014-05-20 21:56:42 | | 6300000001 | 2014-05-20 21:56:42 | | 6750000001 | 2014-05-20 21:56:42 | | 5310000001 | 2014-05-20 21:56:42 | | 7200000001 | 2014-05-20 21:56:42 | | 7380000001 | 2014-05-20 21:56:42 | | 5760000001 | 2014-05-20 21:56:42 | | 6930000001 | 2014-05-20 21:56:42 | | 6660000001 | 2014-05-20 21:56:42 | | 5670000001 | 2014-05-20 21:56:42 | | 6210000001 | 2014-05-20 21:56:42 | | 5850000001 | 2014-05-20 21:56:42 | | 6570000001 | 2014-05-20 21:56:42 | | 5580000001 | 2014-05-20 21:56:42 | | 5130000001 | 2014-05-20 21:56:42 | | 7290000001 | 2014-05-20 21:56:42 | | 6390000001 | 2014-05-20 21:56:42 | | 5490000001 | 2014-05-20 21:56:42 | | 5220000001 | 2014-05-20 21:56:42 | | 7560000001 | 2014-05-20 21:56:42 | | 7470000001 | 2014-05-20 21:56:42 | | 7020000001 | 2014-05-20 21:56:42 | | 6840000001 | 2014-05-20 21:56:42 | | 6030000001 | 2014-05-20 21:56:42 | | 6480000001 | 2014-05-20 21:56:42 | | 7650000001 | 2014-05-20 21:56:42 | | 5940000001 | 2014-05-20 21:56:42 | | 6120000001 | 2014-05-20 21:56:42 | | 7740000001 | 2014-05-20 21:56:42 | | 5400000001 | 2014-05-20 21:56:42 | | 5760000001 | 2014-05-21 03:19:07 | | 6840000001 | 2014-05-21 03:19:17 | | 7020000001 | 2014-05-21 03:19:32 | | 7200000001 | 2014-05-21 03:19:45 | | 7110000001 | 2014-05-21 03:19:46 | | 7380000001 | 2014-05-21 03:19:48 | | 5670000001 | 2014-05-21 03:19:58 | | 6930000001 | 2014-05-21 03:19:59 | | 6030000001 | 2014-05-21 03:20:00 | | 5940000001 | 2014-05-21 03:20:00 | | 7290000001 | 2014-05-21 03:20:02 | | 6120000001 | 2014-05-21 03:20:09 | | 5850000001 | 2014-05-21 03:20:18 | | 5580000001 | 2014-05-21 03:20:24 | | 6480000001 | 2014-05-21 03:25:05 | | 6390000001 | 2014-05-21 03:25:37 | | 6210000001 | 2014-05-21 03:25:45 | | 7470000001 | 2014-05-21 03:26:14 | | 6750000001 | 2014-05-21 03:27:17 | | 5310000001 | 2014-05-21 03:27:33 | | 5130000001 | 2014-05-21 03:27:34 | | 6570000001 | 2014-05-21 03:27:34 | | 7560000001 | 2014-05-21 03:27:45 | | 5220000001 | 2014-05-21 03:27:45 | | 5400000001 | 2014-05-21 03:27:53 | | 5490000001 | 2014-05-21 03:27:55 | | 6660000001 | 2014-05-21 03:28:07 | | 6300000001 | 2014-05-21 03:28:13 | | 7740000001 | 2014-05-21 03:28:26 | | 7650000001 | 2014-05-21 03:28:37 | +------------+---------------------+60 rows in set (0.00 sec)

接下来，我们要检索所有IDS的平均时间。 有以下两种方式：

第一， 对表进行了两次访问，并且有GROUP BY 操作，不可取。

mysql> select sec_to_time(avg(timestampdiff(second,a.times,b.times)))  as 'running'     -> from     -> (select ids,min(log_time) as times from log_ytt where 1 group by ids ) as a,    -> (select ids,max(log_time) as times from log_ytt where 1 group by ids) as b where a.ids = b.ids;+---------------+| running       |+---------------+| 05:27:08.8333 | +---------------+1 row in set (0.00 sec)

第二，虽然对表进行了最少的访问，但是也有一次GROUP BY 操作。也没办法，表设计如此。

mysql> SELECT SEC_TO_TIME(AVG(times)) AS 'Running' FROM     -> (    -> SELECT TIMESTAMPDIFF(SECOND,MIN(log_time),MAX(log_time)) AS times FROM log_ytt GROUP BY ids    -> ) AS T;+---------------+| Running       |+---------------+| 05:27:08.8333 | +---------------+1 row in set (0.00 sec)

宽表：log_ytt_horizontal.

mysql> show create table log_ytt_horizontal;+------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+| Table                  | Create Table                                                                                                                                                                                              |+------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+| log_ytt_horizontal | CREATE TABLE `log_ytt_horizontal` (  `ids` bigint(20) NOT NULL,  `start_time` datetime DEFAULT NULL,  `end_time` datetime DEFAULT NULL,  PRIMARY KEY (`ids`)) ENGINE=InnoDB DEFAULT CHARSET=utf8 | +------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+1 row in set (0.00 sec)

表记录数：

mysql> select * from log_ytt_horizontal;+------------+---------------------+---------------------+| ids        | start_time          | end_time            |+------------+---------------------+---------------------+| 5130000001 | 2014-05-20 21:56:42 | 2014-05-21 03:27:34 | | 5220000001 | 2014-05-20 21:56:42 | 2014-05-21 03:27:45 | | 5310000001 | 2014-05-20 21:56:42 | 2014-05-21 03:27:33 | | 5400000001 | 2014-05-20 21:56:42 | 2014-05-21 03:27:53 | | 5490000001 | 2014-05-20 21:56:42 | 2014-05-21 03:27:55 | | 5580000001 | 2014-05-20 21:56:42 | 2014-05-21 03:20:24 | | 5670000001 | 2014-05-20 21:56:42 | 2014-05-21 03:19:58 | | 5760000001 | 2014-05-20 21:56:42 | 2014-05-21 03:19:07 | | 5850000001 | 2014-05-20 21:56:42 | 2014-05-21 03:20:18 | | 5940000001 | 2014-05-20 21:56:42 | 2014-05-21 03:20:00 | | 6030000001 | 2014-05-20 21:56:42 | 2014-05-21 03:20:00 | | 6120000001 | 2014-05-20 21:56:42 | 2014-05-21 03:20:09 | | 6210000001 | 2014-05-20 21:56:42 | 2014-05-21 03:25:45 | | 6300000001 | 2014-05-20 21:56:42 | 2014-05-21 03:28:13 | | 6390000001 | 2014-05-20 21:56:42 | 2014-05-21 03:25:37 | | 6480000001 | 2014-05-20 21:56:42 | 2014-05-21 03:25:05 | | 6570000001 | 2014-05-20 21:56:42 | 2014-05-21 03:27:34 | | 6660000001 | 2014-05-20 21:56:42 | 2014-05-21 03:28:07 | | 6750000001 | 2014-05-20 21:56:42 | 2014-05-21 03:27:17 | | 6840000001 | 2014-05-20 21:56:42 | 2014-05-21 03:19:17 | | 6930000001 | 2014-05-20 21:56:42 | 2014-05-21 03:19:59 | | 7020000001 | 2014-05-20 21:56:42 | 2014-05-21 03:19:32 | | 7110000001 | 2014-05-20 21:56:42 | 2014-05-21 03:19:46 | | 7200000001 | 2014-05-20 21:56:42 | 2014-05-21 03:19:45 | | 7290000001 | 2014-05-20 21:56:42 | 2014-05-21 03:20:02 | | 7380000001 | 2014-05-20 21:56:42 | 2014-05-21 03:19:48 | | 7470000001 | 2014-05-20 21:56:42 | 2014-05-21 03:26:14 | | 7560000001 | 2014-05-20 21:56:42 | 2014-05-21 03:27:45 | | 7650000001 | 2014-05-20 21:56:42 | 2014-05-21 03:28:37 | | 7740000001 | 2014-05-20 21:56:42 | 2014-05-21 03:28:26 | +------------+---------------------+---------------------+30 rows in set (0.00 sec)

如果对这种稍微冗余一些的表来进行查询，那么对表的访问以及CPU的资源占用都达到了最低。

mysql> select sec_to_time(avg(timestampdiff(second,start_time,end_time))) as 'Running'  from log_ytt_horizontal;+---------------+| Running       |+---------------+| 05:27:08.8333 | +---------------+1 row in set (0.00 sec)