SQL优化 - hint - driving_site

driving_site用于分布式查询中，指定数据集发送到那个数据库上执行。在某些情况下可以大大提高SQL的性能。下面是一个小测试：

1. 在远程数据库上，创建测试表test_remote

pd@DWTEST>create table test_remote 2 as 3 select rownum rn, a.* from user_objects a; Table created. pd@DWTEST>insert into test_remote select * from test_remote; 10 rows created. pd@DWTEST>/ 20 rows created. pd@DWTEST>/ 40 rows created. pd@DWTEST>/ ... 655360 rows created. pd@DWTEST>update test_remote set rn = rownum; 1310720 rows updated. pd@DWTEST>commit; Commit complete.

2. 在本地数据库，创建测试表test_local：

C:\Documents and Settings\yuechao.tianyc>sqlplus test/test SQL*Plus: Release 10.2.0.1.0 - Production on 星期三 4月 29 14:37:24 2009 Copyright (c) 1982, 2005, Oracle. All rights reserved. 连接到: Oracle Database 10g Enterprise Edition Release 10.2.0.1.0 - Production With the Partitioning, OLAP and Data Mining options SQL> drop table test_local purge; 表已删除。 SQL> create table test_local 2 as 3 select rownum rn, a.* from user_objects a; 表已创建。 SQL> insert into test_local select * from test_local; 已创建48行。 SQL> / ... 已创建768行。 SQL> update test_local set rn = rownum; 已更新1536行。 SQL> commit; 提交完成。

3. 通过driving_site，比较数据在远程和本地执行速度的不同：

-- 1. 直接执行SQL，耗时0.93m，通过执行计划发现是将远程表test_remote拉到本地后执行hash join的。 SQL> set timing on SQL> set linesize 1000 SQL> set pagesize 100 SQL> explain plan for 2 select count(*) from test_local l, test_remote@to_s12 r 3 where l.rn = r.rn; 已解释。 已用时间: 00: 00: 00.00 SQL> select count(*) from test_local l, test_remote@to_s12 r 2 where l.rn = r.rn; COUNT(*) ---------- 1536 已用时间: 00: 00: 00.93 SQL> select * from table(dbms_xplan.display); PLAN_TABLE_OUTPUT --------------------------------------------------------------------------------------------------- Plan hash value: 2814429697 --------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Inst |IN-OUT| --------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 26 | 10 (10)| 00:00:01 | | | | 1 | SORT AGGREGATE | | 1 | 26 | | | | | |* 2 | HASH JOIN | | 327 | 8502 | 10 (10)| 00:00:01 | | | | 3 | REMOTE | TEST_REMOTE | 327 | 4251 | 2 (0)| 00:00:01 | TO_S12 | R->S | | 4 | TABLE ACCESS FULL| TEST_LOCAL | 1536 | 19968 | 7 (0)| 00:00:01 | | | --------------------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 2 - access("L"."RN"="R"."RN") Remote SQL Information (identified by operation id): ---------------------------------------------------- 3 - SELECT "RN" FROM "TEST_REMOTE" "R" (accessing 'TO_S12.REGRESS.RDBMS.DEV.US.ORACLE.COM' ) Note ----- - dynamic sampling used for this statement 已选择27行。 已用时间: 00: 00: 00.01 -- 2. 通过driving_site，将本地表test_local发送到远程执行，再将结果集返回本地。耗时0.34m SQL> select/*+driving_site(r)*/ count(*) from test_local l, test_remote@to_s12 r 2 where l.rn = r.rn; COUNT(*) ---------- 1536 已用时间: 00: 00: 00.34 SQL> explain plan for 2 select/*+driving_site(r)*/ count(*) from test_local l, test_remote@to_s12 r 3 where l.rn = r.rn; 已解释。 已用时间: 00: 00: 00.14 SQL> select * from table(dbms_xplan.display); PLAN_TABLE_OUTPUT ------------------------------------------------------------------------------------------------------- Plan hash value: 3396146028 ------------------------------------------------------------------------------------------------------ | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Inst |IN-OUT| ------------------------------------------------------------------------------------------------------ | 0 | SELECT STATEMENT REMOTE| | 1 | 26 | 4970 (1)| 00:01:00 | | | | 1 | SORT AGGREGATE | | 1 | 26 | | | | | |* 2 | HASH JOIN | | 327 | 8502 | 4970 (1)| 00:01:00 | | | | 3 | REMOTE | TEST_LOCAL | 327 | 4251 | 3 (0)| 00:00:01 | ! | R->S | | 4 | TABLE ACCESS FULL | TEST_REMOTE | 1130K| 14M| 4961 (1)| 00:01:00 | DWTEST | | ------------------------------------------------------------------------------------------------------ Predicate Information (identified by operation id): --------------------------------------------------- 2 - access("A2"."RN"="A1"."RN") Remote SQL Information (identified by operation id): ---------------------------------------------------- 3 - SELECT "RN" FROM "TEST_LOCAL" "A2" (accessing '!' ) Note ----- - fully remote statement - dynamic sampling used for this statement 已选择27行。 已用时间: 00: 00: 00.01

4. 通过上面的测试，可以发现二者的执行时间是不一样的。我们再将二者分别执行100次，比较一下平均时间：

-- 将远程表拉到本地执行，耗时65.71（单位为1/100秒） SQL> set serveroutput on SQL> declare 2 n_count number; 3 n_begin_time number; 4 n_sum_time number; 5 begin 6 n_sum_time := 0; 7 for n_loop in 1..100 loop 8 n_begin_time := dbms_utility.get_cpu_time; 9 select count(*) 10 into n_count 11 from test_local l, test_remote@to_s12 r 12 where l.rn = r.rn; 13 n_sum_time := n_sum_time + (dbms_utility.get_cpu_time - n_begin_time); 14 end loop; 15 dbms_output.put_line('avg cpu_time:'||(n_sum_time/100)); 16 end; 17 / avg cpu_time:65.71 PL/SQL 过程已成功完成。 已用时间: 00: 01: 28.39 -- 将本地表发送到远程执行，再将结果返回到本地，耗时0.05（单位为1/100秒） SQL> declare 2 n_count number; 3 n_begin_time number; 4 n_sum_time number; 5 begin 6 n_sum_time := 0; 7 for n_loop in 1..100 loop 8 n_begin_time := dbms_utility.get_cpu_time; 9 select/*+driving_site(r)*/ count(*) 10 into n_count 11 from test_local l, test_remote@to_s12 r 12 where l.rn = r.rn; 13 n_sum_time := n_sum_time + (dbms_utility.get_cpu_time - n_begin_time); 14 end loop; 15 dbms_output.put_line('avg cpu_time:'||(n_sum_time/100)); 16 end; 17 / avg cpu_time:.05 PL/SQL 过程已成功完成。 已用时间: 00: 00: 23.14

5. 结论

在分布式查询中，当一张表比较小，而且最终得到的结果集也比较小的话，使用driving_site将小表发送到大表端执行是比较快的。