driving_site用於分佈式查詢中,指定數據集發送到那個數據庫上執行。在某些情況下可以大大提高SQL的性能。下面是一個小測試:
1. 在遠程數據庫上,創建測試表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. 在本地數據庫,創建測試表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.
連接到:
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;
表已刪除。
SQL> create table test_local
2 as
3 select rownum rn, a.* from user_objects a;
表已創建。
SQL> insert into test_local select * from test_local;
已創建48行。
SQL> /
...
已創建768行。
SQL> update test_local set rn = rownum;
已更新1536行。
SQL> commit;
提交完成。
|
3. 通過driving_site,比較數據在遠程和本地執行速度的不同:
-- 1. 直接執行SQL,耗時0.93m,通過執行計劃發現是將遠程表test_remote拉到本地後執行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;
已解釋。
已用時間: 00: 00: 00.00
SQL> select count(*) from test_local l, test_remote@to_s12 r
2 where l.rn = r.rn;
COUNT(*)
----------
1536
已用時間: 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
已選擇27行。
已用時間: 00: 00: 00.01
-- 2. 通過driving_site,將本地表test_local發送到遠程執行,再將結果集返回本地。耗時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
已用時間: 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;
已解釋。
已用時間: 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
已選擇27行。
已用時間: 00: 00: 00.01
|
4. 通過上面的測試,可以發現二者的執行時間是不一樣的。我們再將二者分別執行100次,比較一下平均時間:
-- 將遠程表拉到本地執行,耗時65.71(單位爲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 過程已成功完成。
已用時間: 00: 01: 28.39
-- 將本地表發送到遠程執行,再將結果返回到本地,耗時0.05(單位爲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 過程已成功完成。
已用時間: 00: 00: 23.14
|
5. 結論
在分佈式查詢中,當一張表比較小,而且最終得到的結果集也比較小的話,使用driving_site將小表發送到大表端執行是比較快的。