编程生活

   :: 首页 :: 新随笔 :: 联系 :: 聚合  :: 管理 ::
  113 随笔 :: 0 文章 :: 18 评论 :: 0 Trackbacks

Understanding Rules-of-Thumb

The syntax of SQL statements has a significant affect on performance. The use of certain command clauses can disable indexes or cause inefficient data sorting and filtering. In some cases, the order in which command clauses are used or the order in which data and tables are referenced can add an extra burden on resources.

Oracle SQL Analyze contains a set of rules, developed by database experts, that evaluates SQL statements and suggests alternative statements, when possible. These rules focus on principles of optimizing performance, such as:

  • enabling indexes to eliminate the need for full table scans
  • reducing the number of sorts, merges, and filtering operations required
  • reducing the number of rows that need to be sorted, filtered, or merged

Oracle SQL Analyze applies these "rules-of-thumb" when you tune a statement with the Tuning Wizard, and supplies alternative SQL statements when possible. Oracle SQL Analyze checks your statement against the following rules, which are explained in this section:

  • Use NOT EXISTS instead of NOT IN
  • Use NOT EXISTS or NOT IN with hints instead of MINUS
  • Use TRUNC differently to enable indexes
  • Use operators differently to enable indexes
  • Do not use columns on both sides of operator
  • Use WHERE in place of HAVING
  • Use UNION ALL instead of UNION

Use NOT EXISTS instead of NOT IN

Using NOT EXISTS instead of NOT IN adds a limiting condition to your queries that can reduce the number of full table scans necessary.

The following example uses a NOT IN clause to find names and department IDs in the DEPARTMENT table where the department ID does not also exist in the EMPLOYEE table:

SELECT name, department_id
FROM department
WHERE department_id NOT IN
(SELECT department_id FROM employee)

Because NOT IN does not use a limiting condition, Oracle will perform a full table scan of DEPARTMENT. For each record in DEPARTMENT, the subquery will be executed. Since the subquery has no limiting WHERE clause, it will perform a full table scan for every record in the full table scan of DEPARTMENT.

Instead, NOT EXISTS can be used so that nested index scans will be used in the subquery for each row in the DEPARTMENT table. The logic of the NOT EXISTS clause tells Oracle not to return the row if it finds a match in both tables. The only records that will be returned from DEPARTMENT are those that return no rows from the subquery, and no full table scans are performed by the subquery. The following statement, therefore, is more efficient than the previous example.

SELECT name, department_id
FROM department,
WHERE NOT EXISTS
(SELECT department_id
FROM employee
WHERE department.department_id=employee.department_id)

Use NOT EXISTS or NOT IN with hints instead of MINUS

MINUS returns the set of rows from one query that is not present in the set of rows returned by a second query. Rewriting queries using NOT EXISTS or NOT IN can enable them to take advantage of indexes, reducing the number of full table scans a clause may require.

In some cases, Oracle SQL Analyze might determine that because a hash anti-join (HASH_AJ) usually does not require a sort, it will produce better results than MINUS.

The following query, for example, matches names and birthdates in the EMPLOYEE table with those in the STOCKHOLDER table, then returns the names and birthdates of employees who are not stockholders. Because MINUS does not use indexes, Oracle will use two full table scans and perform a sort on each table before the MINUS operation can be performed.

SELECT birth_date, last_name, first_name
FROM employee
MINUS
SELECT birth_date, last_name, first_name
FROM stock_holder

If the statement is re-written using NOT EXISTS, Oracle can use nested index scans in the subquery for rows in the primary statement.

SELECT birth_date, last_name, first_name
FROM employee
WHERE NOT EXISTS
(SELECT 1
  FROM stock_holder
WHERE stock_holder.birth_date = employee.birth_date
  AND stock_holder.first_name = employee.first_name)

If Oracle SQL Analyze determines that a hash anti-join will produce better results, the example query could be rewritten to use two full table scans and an anti-join algorithm to join the rows, instead of performing sort and minus operations.

SELECT birth_date, last_name, first_name
FROM employee
WHERE (birth_date, last_name, first_name)NOT IN
(SELECT /*+ hash_aj (stock_holder) */ birth_date, last_name, first_name
FROM stock_holder)

Use TRUNC differently to enable indexes

Using the truncate command (TRUNC) on an indexed column disables the index. Rewriting your query so that fewer columns are truncated allows it to take advantage of indexes to increase performance.

In the following example, trans_date is an indexed column, but the index is disabled by the TRUNC command.


SELECT account_name, trans_date
FROM transaction
WHERE TRUNC(trans_date) = TRUNC(sysdate)

The query can be rewritten as shown below to use the trans_date index and increase performance.

SELECT account_name, trans_date
FROM transaction
WHERE trans_date BETWEEN TRUNC(sysdate) AND TRUNC(sysdate) + .99999

Use operators differently to enable indexes

The optimizer does not use an index if the indexed column is part of a function (in the WHERE clause). If Oracle SQL Analyze determines that an equation can be rewritten to avoid the use of operators, it can rewrite the statement as shown below.

In this example, the equation in the query can be rewritten as a simple inequality clause. statement. Therefore the query

SELECT account_name, trans_date, amount
FROM transaction
WHERE amount + 3000 < 5000

can be rewritten as


SELECT account_name, trans_date, amount
FROM transaction
WHERE amount < 2000

Do not use columns on both sides of operator

When an indexed column appears on both sides of an operator, the index for that column is disabled. Oracle SQL Analyze detects this condition and, when possible, rewrites the statement to allow the index to be used.

In the following example, the column account_name is indexed, but the index is disabled.

SELECT account_name, trans_date, amount
FROM transaction
WHERE account_name = NVL(:acc_name, account_name)

The query can be rewritten using LIKE so that the indexed column is only on one side of the operator.

SELECT account_name, trans_date, amount
FROM transaction
WHERE account_name LIKE NVL(:acc_name, `%')

Use WHERE in place of HAVING

The HAVING clause limits rows collected by a GROUP BY clause only after they have been aggregated. Whenever possible, it is better to limit the number of rows retrieved before they are merged and sorted into an aggregation. Using WHERE in place of HAVING eliminates rows before they are added to the aggregation.

The statement below sorts an entire list of items by quantity, then removes from the aggregation all items with a quantity less than 40.

SELECT quantity, AVG(actual_price)
FROM item
GROUP BY quantity
HAVING quantity > 40

The statement can be rewritten so that all rows where QUANTITY is less than 40 are removed before the aggregation is sorted.

SELECT quantity, AVG(actual_price)
FROM item
WHERE quantity >40
GROUP BY quantity

Note that if the HAVING clause is applied to aggregate functions, it cannot be replaced by WHERE. In the query below, for example, HAVING is applied to a SUM function.

SELECT program_name
      ,count
      ,min(end_date-start_date) "Min Runtime"
      ,avg(end_date-start_date)"Avg Runtime"
      ,max((end_date-start_date)"Max Runtime"
      ,sum(end_date-start_date)"tot Runtime"
FROM jobs
WHERE start_date>sys_date - 7
GROUP BY program_name
HAVING sum((end_date-start_date)>0.25 or max(end_date-start_date) > 0.04

Use UNION ALL instead of UNION

The difference between the UNION and UNION ALL is that UNION requires a sort operation to eliminate any rows that are duplicated across the two row sets, while UNION ALL returns all rows, even if they are duplicated. If duplicated rows are not important, using UNION ALL can avoid potentially expensive sorts, merges, and filtering operations.

For example, the statement

SELECT acct_num, balance_amt
FROM debit_transactions
WHERE tran_date = `31-DEC-99'
UNION
SELECT acct_num, balance_amt
FROM credit_transactions
WHERE tran_date = `31-DEC-99'

Can be rewritten as

SELECT acct_num, balance_amt
FROM debit_transactions
WHERE tran_date = `31-DEC-99'
UNION ALL
SELECT acct_num, balance_amt
FROM credit_transactions
WHERE tran_date = `31-DEC-99'

Using the SQL Tuning Wizard

The SQL Tuning Wizard guides you through the entire SQL statement tuning process. It evaluates your SQL Statement using Rules-of-Thumb to generate alternate, optimized versions of your SQL statement.

To use the SQL Tuning Wizard:

Select Tools=>SQL Tuning Wizard. This launches the SQL Tuning Wizard.

The SQL Tuning Wizard Process

The SQL Tuning Wizard is an automated guide that leads you through tuning a SQL statement. Throughout the process, you will be able to make choices that will help the wizard optimize your specific SQL statement. If you need more information to make your choices, select the Help button from any of the wizard pages.

The SQL Tuning Wizard will guide you through the following processes:

  • Evaluation

    The evaluation process identifies inefficiencies in the way that your SQL statement is written. The SQL Tuning Wizard provides a graph with a projected improvement percentage that is based on a modified version of the SQL.

    The SQL Tuning Wizard projected improvement graph is derived from the information collected by the system optimizer. In some cases the SQL Tuning Wizard may detect inefficiencies in the way that the statement was written, but may not be able to predict the degree of performance improvement. It may still be worthwhile to look at the modified SQL statement, however, to see if the changes have improved the overall performance of the SQL statement.

  • Recommendations

    The recommendation review process allows you to see which rules have been violated by the SQL statement. For each rule that is checked, the SQL Tuning Wizard provides a recommendation that improves the SQL statement. You can also view the Rule Details for each of the rules listed. You may chose to accept (checked) or decline (unchecked) recommendations for any of the rules.

    By default, a rule is checked only if the recommendation is guaranteed to return the same result set as the original SQL statement.

  • Explain Plan Comparison

    The comparison process allows you to compare the original SQL statement to the modified statement to verify the actual performance improvements for all of the recommendations you accepted. You can compare the changes to the actual SQL statements before choosing to accept the modified statements. Once you have verified the performance improvements, you can Execute the modified SQL statements.

Using the Hint Wizard

The Hint Wizard identifies hints in a statement and allows the user to present other hints that can be added to the statement. It provides a description for a selected hint and will automatically generate a new SQL statement if a hint is added or deleted.

To use the Hint Wizard:

Select Tools=>Hint Wizard. The Hint Wizard will guide you through the rest of this process.

  1. Select a subquery to analyze from the Hint Wizard page.
  2. View/delete the current hints.
  3. Select a new hint to add, and supply:
    • table parameters, if necessary.
    • index parameters, if necessary.
  4. Review the current hints.
  5. Apply hints to the SQL statement.
posted on 2007-10-25 20:54 wilesun 阅读(274) 评论(0)  编辑  收藏

只有注册用户登录后才能发表评论。


网站导航: