avh新一代智能SQL语句优化技术课件.ppt

DM10:Sybase SQL Expert 12.5新一代智能SQL语句优化技术The New Generation of Intelligent SQL Tuning Technology,杜伟业 ( Richard To) R技术总监 灵高科研有限公司,什么是反应时间,1/10秒是用户认为系统能够立即反应的极限,1 秒是用户觉得没有被中断的极限,10秒是用户能将注意力继续集中在与计算机的对话上的时间极限,引用自Jakob Nielsen 的 “Usability Engineering” 第五章(Morgan Kaufmann, San Francisco出版),我们可以关注的地方,可调整的部分:硬件操作系统数据库网络应用程序,许多人在试图提高数据库应用程序性能时只将注意力放在了前三项：硬件、操作系统、数据库,数据库系统各部分对性能的影响,应用程序,数据库,操作系统,硬件,网络,Wrong Focus,应用程序,其它,最 少 60% 的 数 据 库 性 能 问 题 是 由 应 用 程 序 引 起 的,性能的挑战,我们可以优化什么?,数据库设计问题编程问题索引问题SQL问题,什么是 SQL ? (结构化查询语言),访问和更新数据对应用程序的性能产生最重大的影响SQL语句消耗70%到90%的数据库系统资源SQL语句可以有多种写法易学，但难以精通需要考虑在数据和系统环境变化后仍能达到最佳化,数据库是如何处理SQL语句的?,SQL,数据库的优化器是如何工作的?,内部重写并产生多种执行计划,执行计划1,执行计划2,执行计划3,成本估计,执行计划1 成本=1000,执行计划2 成本=3000,执行计划3 成本= 500,执行,优化器说明,什么是执行计划SQL语法如何影响优化器的决定,什么是SQL优化?,解决这个问题的传统途径,执行计划,SET SHOWPLAN ONGOSELECT * FROM EMPLOYEE WHERE EMP_ID 73712GOSET SHOWPLAN OFFGO,QUERY PLAN FOR STATEMENT 1 (at line 1). STEP 1 The type of query is SELECT. FROM TABLE EMPLOYEE Nested iteration. Table Scan. Forward scan. Positioning at start of table. Using I/O Size 2 Kbytes for data pages. With LRU Buffer Replacement Strategy for data pages.Total estimated I/O cost for statement 1 (at line 1): 20600.,iSQL,IO和时间统计信息,SET STATISTICS IO ONGOSET STATISTICS TIME ONGOSELECT * FROM EMPLOYEE WHERE EMP_ID 73712GOSET STATISTICS IO OFFGOSET STATISTICS TIME OFFGO,iSQL,Server Message: Number 3631, Severity 10Line 1:Total actual I/O cost for this command: 0. Total writes for this command: 0 Server Message: Number 3631, Severity 10Line 1:Total actual I/O cost for this command: 0. Total writes for this command: 0 Execution Time 0. SQL Server cpu time: 0 ms. SQL Server elapsed time: 3614 ms.Parse and Compile Time 0. SQL Server cpu time: 0 ms.Table: EMPLOYEE scan count 1, logical reads: (regular=1030 apf=0 total=1030), physical reads: (regular=0 apf=0 total=0), apf IOs used=0 Server Message: Number 3631, Severity 10Line 1:Total actual I/O cost for this command: 2060. Total writes for this command: 0 Execution Time 16. SQL Server cpu time: 1600 ms. SQL Server elapsed time: 7730 ms.(18371 rows affected)Parse and Compile Time 0. SQL Server cpu time: 0 ms. Execution Time 0. SQL Server cpu time: 0 ms. SQL Server elapsed time: 0 ms.Parse and Compile Time 0. SQL Server cpu time: 0 ms.,Traceon,DBCC TRACEON (3604, 302, 310)GOSELECT * FROM EMPLOYEE WHERE EMP_ID 73712GO,iSQL,Beginning selection of qualifying indexes for table EMPLOYEE,varno = 0, objectid 1593824790. The table (Allpages) has 18373 rows, 1030 pages, Data Page Cluster Ratio 0.998890Table scan cost is 18373 rows, 1030 pages, using no data prefetch (size 2K I/O), in data cache default data cache (cacheid 0) with LRU replacementSelecting best index for the SEARCH CLAUSE:EMPLOYEE.EMP_ID 73712.000000Estimated selectivity for EMP_ID, selectivity = 0.999946.Estimating selectivity of index EMPLOYEE_1961957491, indid 1 scan selectivity 0.999946, filter selectivity 0.999946 18372 rows, 1032 pages, index height 2, Data Row Cluster Ratio 1.000000, Index Page Cluster Ratio 0.000000, Data Page Cluster Ratio 0.998890.,使用 EXIST 还是 IN？ SELECT * FROM A WHERE A.CITY IN (SELECT B.CITY FROM B)SELECT * FROM A WHERE EXISTS (SELECT X FROM B WHERE B.CITY = A.CITY),EXIST or IN,表连接与子查询,通情况下表连接(Table join)要好过子查询(sub-query)源语句 SELECT * FROM A WHERE A.CITY IN (SELECT B.CITY FROM B)重写后SELECT A.* FROM A, B WHERE A.CITY = B.CITY /* if B.CITY is unique in B */,如果你确实知道问题所在，可以强制使用一个特殊的表操纵路径( driving path )。源语句SELECT * FROM A, B WHERE A.STATE = B.STATE重写后的语句/* 禁止使用B的索引来改变表操纵路径 (B drive A) */SELECT * FROM A, B WHERE A.STATE = ISNULL(B.STATE,B.STATE),Joins Change Table Driving Path,要使用索引就不能在该索引字段上进行操作源语句 /* A.key1上的索引被使用*/SELECT * FROM A WHERE A.Key1 = ver1 AND A.Key2 = ver2重写后 /* 禁止使用字段A.key1上的索引，使用字段A.Key2上的索引*/SELECT * FROM A WHERE ISNULL(A.Key1, A.key1) = ver1 AND A.Key2 = ver2,Use Indexes,学习使用强制选项(Forces Options) SET FORCEPLAN ONINDEX hintPARALLEL hintSET SORT_MERGE ONSET JTC ONSET TABLE COUNT,Forces,使用FORCEPLAN 强制选项强制使用特殊的表操纵路径 (driving path)。原来的方法 /*禁止使用B的索引来改变表操纵路径 (B drive A)*/SELECT * FROM A, B WHERE A.STATE = ISNULL(B.STATE,B.STATE)用 FORCEPLAN重写后SET FORCEPLAN ONGOSELECT * FROM B, A WHERE A.STATE = B.STATEGO,Joins Change Table Driving Path,强制索引(Force Index),SELECT * FROM EMPLOYEE WHERE EMP_ID 73712,Force Index,备注：仅在事先知道使用使用索引 x 将会较快的情况下采用,(INDEX EMP_ID_INX),QUERY PLAN FOR STATEMENT 1 (at line 1). STEP 1 The type of query is SELECT. FROM TABLE EMPLOYEE Nested iteration. Table Scan. Forward scan. Positioning at start of table. Using I/O Size 2 Kbytes for data pages. With LRU Buffer Replacement Strategy for data pages.Total estimated I/O cost for statement 1 (at line 1): 20600.,QUERY PLAN FOR STATEMENT 1 (at line 1). STEP 1 The type of query is SELECT. FROM TABLE EMPLOYEE Nested iteration. Index : EMP_ID_INX Forward scan. Positioning by key. Keys are: EMP_ID ASC Using I/O Size 2 Kbytes for index leaf pages. With LRU Buffer Replacement Strategy for index leaf pages. Using I/O Size 2 Kbytes for data pages. With LRU Buffer Replacement Strategy for data pages.Total estimated I/O cost for statement 1 (at line 1): 23382.,变量使用上的普遍问题,SELECT * FROM employee WHERE (emp_id range_lower OR range_lower IS NULL) AND (emp_id range_upper OR range_upper IS NULL),使用配合输入范围边界的索引范围搜索(index range search),SELECT * FROM employee WHERE emp_id ISNULL(range_lower , min of emp_id) AND emp_id ISNULL(range_upper, max of emp_id),SELECT * FROM A WHERE EXISTS (SELECT x FROM B WHERE A.key = B.key),SELECT * FROM A a1 WHERE EXISTS IN (SELECT x FROM A a2, B WHERE a1.key = B.key AND a1.Unique_Key = a2.Unique_Key),复杂的SQL转化,语义上等同于:,这条SQL语句有多少个不同的等价写法?,SELECT * FROM DEPARTMENT WHERE DPT_ID IN (SELECT EMP_DEPT FROM EMPLOYEE WHERE EMP_ID 50000),更多的信息：DEPARTMENT EMPLOYEE记录数 = 410记录数 = 18370索引 - DPT_ID索引 - EMP_ID,最新的SQL优化技术,基于人工智能(A.I.)的 SQL 转换技术使用人工智能(A.I.)来重写 SQL语句,ESR 人工智能引擎,优化的 SQL 1,优化的 SQL 2,优化的 SQL 3,优化的 SQL 4,优化的 SQL 5,优化的 SQL 6,源代码,Sybase SQL Expert 12.5 使用的技术,不良的SQL,最好的 SQL,外部SQL重写(External SQL Rewriter -ESR),人工智能( A.I.) SQL转换技术,优化后的 SQL,SQL,Feedback Searching Engine反馈式搜索引擎,什么是递归的 SQL 转换 (Recursive SQL Transformation)?,SELECT * FROM AWHERE A.C1 IN (SELECT B.C1 FROM B WHERE EXISTS (SELECT x FROM C WHERE B.C2=C.C2 ),IN to EXISTS,EXISTS to IN,SELECT * FROM A WHERE EXISTS (SELECT x FROM B WHERE EXISTS (SELECT x FROM C WHERE B.C2=C.C2) WHERE A.C1=B.C1),SELECT * FROM A WHERE A.C1 IN (SELECT B.C1 FROM B WHERE B.C2 IN (SELECT C.C2 FROM C),IN to EXISTS,EXISTS to IN,SELECT * FROM A WHERE A.C1 IN (SELECT B.C1 FROM B WHERE EXISTS (SELECT x FROM C WHERE B.C2=C.C2 ),SELECT * FROM A WHERE EXISTS (SELECT x FROM B WHERE B.C2 IN (SELECT C.C2 FROM C) AND A.C1=B.C1),SELECT * FROM A WHERE A.C1 IN (SELECT B.C1 FROM B WHERE B.C2 IN (SELECT C.C2 FROM C),SELECT * FROM A WHERE EXISTS (SELECT x FROM B WHERE B.C2 IN (SELECT C.C2 FROM C) AND A.C1=B.C1),SELECT * FROM A WHERE A.C1 IN (SELECT B.C1 FROM B WHERE EXISTS (SELECT x FROM C WHERE B.C2=C.C2),SELECT * FROM A WHERE EXISTS (SELECT x FROM B WHERE EXISTS (SELECT x FROM C WHERE B.C2=C.C2) AND A.C1=B.C1),使用2 个转换规则,SQL 优化技术 ESR,SQL1SQL2SQL3SQL4SQL5SQL6SQL7SQL8SQL9SQL10SQL11SQL12SQL.SQLN,人工智能重写 执行计划 性能 结果,SQL,相同的结果,SQL1,SQL2,Which SQL is the Best?,边做边学式训练,演示 2,边做边学式训练SQL 比较器 & 执行计划帮助,Sybase 12.5 最新的技术,Sybase Abstract Plan,什么是 Abstract Plan?,捕获查询的SQL语句并存放该查询的abstract plan在系统表Sysqueryplans中提交给数据库的SQL语句会跟已存于数据库内的SQL语句进行比较，如果匹配，存于数据库的abstract plan 将会用来执行该查询。,Abstract Plan,优化器,执行,基于保存的abstract plan 来产生执行计划,基于 数据库统计信息 来产生执行计划,SQL,Sybase,使用Abstract Plan,/* 创建abstract plan组demo1 */sp_add_qpgroup demo1Go/* 保存abstract plan到组demo1 */set plan dump demo1 onGo/* 执行 SQL */Select * from A, B/* 从组demo1加载计划(Plan) */set plan load demo1 onGoExecution Plan /* 下一次你重新执行同一条SQL语句 */QUERY PLAN FOR STATEMENT 1 (at line 1).Optimized using an Abstract Plan (ID : 1989579095).,保存SQL的计划(Plan),/* 在组 demo1 下创建abstract plan */* 以使用字段A.key2上的索引 */CREATE PLAN“SELECT * FROM A WHERE A.Key1 = 1 AND A.Key2 = 2”“( i_scan Index_key2 A ) ( prop A ( parallel 1 ) ( prefetch 2 ) ( lru ) )” into demo1,使用Abstract Plan的好处,即使没有源代码的情况下也可以对SQL语句进行优化执行计划将不会受以下情况影响：数据库升级引起的数据库内部优化器的改变新的ASE特性调整选项的改变，如：并行深度(parallel degree)，表分区(table partitioning)和索引(indexing),示范,Abstract Plan使用 Abstract Plan优化SQL,使用 Abstract Plan优化SQL语句,SQL Expert,SQL 扫瞄器,. SQL1 SQL2 . SQL3 . SQL4 . SQL5 . SQL6 . SQL7 . SQL8. SQL9 SQL10 . SQL11 . SQL12 SQL13 . SQL14 SQL15 . SQL16 SQL17 SQL18 .,. SQL1 . SQL2 . SQL3 . SQL4 . SQL5 . SQL6 SQL7 . SQL8 . SQL9 SQL10 . SQL11 SQL12 SQL13 . SQL14 . SQL15 SQL16 . .,程序 A,程序 B,SQL1,SQL2,SQL3,SQL4,SQL5,SQL6,SQL7,SQL8,SQL9,SQL10,SQL11,SQL12,SQL13,SQL14,SQL15,SQL16,SQL17,SQL18,SQL1,SQL2,SQL3,SQL4,SQL5,SQL6,SQL7,SQL8,SQL9,SQL10,SQL11,SQL12,SQL13,SQL14,SQL15,SQL16,定位有问题的SQL语句,数据库对象或监视结果,程序源代码,反馈式搜索引擎,优化的 SQL 1,优化的 SQL 2,优化的 SQL 3,优化的 SQL 4,优化的 SQL 5,优化的 SQL 6,演示 3,SQL 扫瞄器定位有问题的SQL语句,高生产力,高质量,高质量 VS 高生产力,谢 谢,杜伟业 ( Richard To) R技术总监 灵高科研有限公司,