Bulk Binds对性能的提高

 当Oracle运行PL/SQL时会使用两套引擎，所有procedural code由PL/SQL engine 完成，所有SQL由SQL engine处理。所以如果Oracle从一个collection中循环执行相同的DML操作，那么为了避免两套engine切换所消耗的系统资源，可以使用bulk binds来把所有的DML操作binding到一次操作中完成。这将极大提高PL/SQL的执行效率。

　　以下是简单的测试，用两种方式插入100000条数据，可以看到效率提高了7倍左右。

SQL> CREATE TABLE test1(
  2    id           NUMBER(10),
  3    description  VARCHAR2(50));

Table created

SQL> ALTER TABLE test1 ADD (
  2    CONSTRAINT test1_pk PRIMARY KEY (id));

Table altered

SQL> SET TIMING ON;

SQL> DECLARE
  2    TYPE id_type          IS TABLE OF test1.id%TYPE;
  3    TYPE description_type IS TABLE OF test1.description%TYPE;
  4
  5    t_id           id_type          := id_type();
  6    t_description  description_type := description_type();
  7  BEGIN
  8    FOR i IN 1 .. 100000 LOOP
  9      t_id.extend;
 10      t_description.extend;
 11
 12      t_id(t_id.last)                   := i;
 13      t_description(t_description.last) := 'Description: ' || To_Char(i);
 14    END LOOP;
 15
 16    FOR i IN t_id.first .. t_id.last LOOP
 17      INSERT INTO test1 (id, description)
 18      VALUES (t_id(i), t_description(i));
 19    END LOOP;
 20
 21    COMMIT;
 22  END;
 23  /

PL/SQL procedure successfully completed

Executed in 141.233 seconds

SQL> truncate table test1;

Table truncated

Executed in 0.631 seconds

SQL>
SQL> DECLARE
  2    TYPE id_type          IS TABLE OF test1.id%TYPE;
  3    TYPE description_type IS TABLE OF test1.description%TYPE;
  4
  5    t_id           id_type          := id_type();
  6    t_description  description_type := description_type();
  7  BEGIN
  8    FOR i IN 1 .. 100000 LOOP
  9      t_id.extend;
 10      t_description.extend;
 11
 12      t_id(t_id.last)                   := i;
 13      t_description(t_description.last) := 'Description: ' || To_Char(i);
 14    END LOOP;
 15
 16    FORALL i IN t_id.first .. t_id.last
 17      INSERT INTO test1 (id, description)
 18      VALUES (t_id(i), t_description(i));
 19
 20    COMMIT;
 21  END;
 22  /

PL/SQL procedure successfully completed

Executed in 27.52 seconds

SQL> select count(*) from test1;

  COUNT(*)
----------
    100000

Executed in 0.04 seconds