该参数只能在当前会话设置。不是oracle端设置的，而是sqlplus设置的，所以我们不能使用设置初始化参数alter system set来改变它，
而是使用set命令，该参数的含义就是sqlplus里预抓取的行数，默认值是15行，也就是服务器进程每次从buffer cache中复制15行到PGA中，从PGA
在发送给客户端。与之相关的另一个参数就是SDU(session data unit)，预抓取在PGA所占的内存大小和SDU大小相关，默认是8K.
SQL> show arraysize
arraysize 15
SQL> create table jason as select * from dba_objects;
Table created.
SQL> select count(1) from dba_objects;
COUNT(1)
———-
50042
SQL> set autot trace
SQL> select * from jason;
50042 rows selected.
Execution Plan
———————————————————-
Plan hash value: 1258222954
—————————————————————————
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
—————————————————————————
| 0 | SELECT STATEMENT | | 44426 | 7679K| 155 (2)| 00:00:02 |
| 1 | TABLE ACCESS FULL| JASON | 44426 | 7679K| 155 (2)| 00:00:02 |
—————————————————————————
Note
—–
– dynamic sampling used for this statement
Statistics
———————————————————-
0 recursive calls
0 db block gets
3992 consistent gets
0 physical reads
0 redo size
2544199 bytes sent via SQL*Net to client
37188 bytes received via SQL*Net from client
3338 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
50042 rows processed
SQL> set arraysize 400
SQL> /
50042 rows selected.
Execution Plan
———————————————————-
Plan hash value: 1258222954
—————————————————————————
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
—————————————————————————
| 0 | SELECT STATEMENT | | 44426 | 7679K| 155 (2)| 00:00:02 |
| 1 | TABLE ACCESS FULL| JASON | 44426 | 7679K| 155 (2)| 00:00:02 |
—————————————————————————
Note
—–
– dynamic sampling used for this statement
Statistics
———————————————————-
0 recursive calls
0 db block gets
817 consistent gets
0 physical reads
0 redo size
1956586 bytes sent via SQL*Net to client
1867 bytes received via SQL*Net from client
127 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
50042 rows processed
可以看到当我们把arraysize从默认的15调整到400后consistent gets逻辑读从原来的3992下降到817，bytes received via SQL*Net from client相关的值也有明显的下降。
arraysize参数的设置对大数据的范围比如全表扫性能提高还是比较大的，所以在网络带宽允许的情况下适当调大arraysize和估算SDU的值，对减少
逻辑读，减少网络交互性，从而介绍CPU的消耗帮助还是比较大的。
