Oracle 11g性能调优实战指南：快速解决数据库问题

4星 · 超过85%的资源需积分: 0 135 浏览量更新于2024-07-27 1 收藏 31.46MB PDF 举报

《Oracle Database 11g Performance Tuning Recipes》是一本面向Oracle数据库管理员的实战指南，旨在解决在实际工作中遇到的性能问题。它采用实例驱动的方法，每个章节聚焦于特定的性能优化领域，提供“菜谱”形式的解决方案，通过具体的例子展示如何执行常见的任务。作者团队深入浅出地解释了背景理论，确保读者在压力环境下也能迅速找到解决之道。本书分为15个章节，内容涵盖了数据库性能优化的关键方面： 1. **第一章：优化表性能** - 着重讨论数据库特性如何影响数据存储和检索的效率，以及在创建表之前需要考虑的物理存储设计。 2. **第二章：选择和优化索引** - 学习如何正确选择和构建索引来加速查询速度，提高查询性能。 3. **第三章：优化实例内存** - 探讨内存管理策略，包括共享池、大池和其他内存组件对性能的影响及优化方法。 4. **第四章：监控系统性能** - 教授如何使用内置工具和技术来实时监测数据库的运行状况，及时发现潜在问题。 5. **第五章：最小化系统竞争** - 解释如何避免多个进程或会话之间的资源争夺，提升并发性能。 6. **第六章：分析操作系统性能** - 针对操作系统层面的性能瓶颈，提供诊断和优化建议。 7. **第七章：数据库故障排除** - 提供故障诊断和修复技巧，确保系统的稳定运行。 8. **第八章：创建高效SQL** - 学习编写和优化SQL语句，提高查询效率。 9. **第九章：手动调优SQL** - 深入探讨高级SQL调优技术，针对特定场景定制优化策略。 10. **第十章：追踪SQL执行** - 教授如何使用追踪工具来识别和理解SQL执行过程中的瓶颈。 11. **第十一章：自动化SQL调优** - 展示如何利用Oracle的自动调优功能进行性能改进，减少人工干预的工作量。 12. **第十二章：执行计划优化与一致性** - 分析执行计划，了解如何确保查询的高效执行。 13. **第十三章：配置优化器** - 讨论如何调整数据库参数以优化查询计划的选择和执行。 14. **第十四章：实施查询提示** - 学习如何使用提示来指导优化器，增强查询性能。 15. **第十五章：并行执行SQL** - 探讨如何利用多处理器优势，提高大规模数据处理的性能。《Oracle Database 11g Performance Tuning Recipes》是一本实战型的性能优化指南，为数据库管理员提供了快速应对性能问题的实用工具和策略，无论是在日常运维还是突发危机情况下都能发挥重要作用。

CHAPTER 1 ■ OPTIMIZING TABLE PERFORMANCE

We realize that to take advantage of this feature the only “solution” is to upgrade to Oracle Database

11g R2, which is oftentimes not an option. However, we felt it was important to discuss this feature

because you’ll eventually encounter the aforementioned characteristics (when you start using the latest

release of Oracle).

You can disable the deferred segment creation feature by setting the database initialization

parameter DEFERRED_SEGMENT_CREATION to FALSE. The default for this parameter is TRUE.

You can also control the deferred segment creation behavior when you create the table. The CREATE

TABLE statement has two new clauses: SEGMENT CREATION IMMEDIATE and SEGMENT CREATION DEFERRED—for

example:

create table f_regs(

reg_id number

,reg_name varchar2(2000))

segment creation immediate;

■ Note The COMPATIBLE initialization parameter needs to be 11.2.0.0.0 or greater before using the SEGMENT

CREATION DEFERRED clause.

1-6. Maximizing Data Loading Speeds

Problem

You’re loading a large amount of data into a table and want to insert new records as quickly as possible.

Solution

Use a combination of the following two features to maximize the speed of insert statements:

• Set the table’s logging attribute to NOLOGGING; this minimizes the generation redo

for direct path operations (this feature has no effect on regular DML operations).

• Use a direct path loading feature, such as the following:

• INSERT /*+ APPEND */ on queries that use a subquery for determining which

records are inserted

• INSERT /*+ APPEND_VALUES */ on queries that use a VALUES clause

• CREATE TABLE…AS SELECT

Here’s an example to illustrate NOLOGGING and direct path loading. First, run the following query to

verify the logging status of a table. In this example, the table name is F_REGS:

CHAPTER 1 ■ OPTIMIZING TABLE PERFORMANCE

select

table_name

,logging

from user_tables

where table_name = 'F_REGS';

Here is some sample output:

TABLE_NAME LOG

------------------------------ ---

F_REGS YES

The prior output verifies that the table was created with LOGGING enabled (the default). To enable

NOLOGGING, use the ALTER TABLE statement as follows:

SQL> alter table f_regs nologging;

Now that NOLOGGING has been enabled, there should be a minimal amount of redo generated for

direct path operations. The following example uses a direct path INSERT statement to load data into the

table:

insert /*+APPEND */ into f_regs

select * from reg_master;

The prior statement is an efficient method for loading data because direct path operations such as

INSERT /*+APPEND */ combined with NOLOGGING generate a minimal amount of redo.

How It Works

Direct path inserts have two performance advantages over regular insert statements:

• If NOLOGGING is specified, then a minimal amount of redo is generated.

• The buffer cache is bypassed and data is loaded directly into the datafiles. This can

significantly improve the loading performance.

The NOLOGGING feature minimizes the generation of redo for direct path operations only. For direct

path inserts, the NOLOGGING option can significantly increase the loading speed. One perception is that

NOLOGGING eliminates redo generation for the table for all DML operations. That isn’t correct. The

NOLOGGING feature never affects redo generation for regular INSERT, UPDATE, MERGE, and DELETE statements.

One downside to reducing redo generation is that you can’t recover the data created via NOLOGGING

in the event a failure occurs after the data is loaded (and before you back up the table). If you can

tolerate some risk of data loss, then use NOLOGGING but back up the table soon after the data is loaded. If

your data is critical, then don’t use NOLOGGING. If your data can be easily re-created, then NOLOGGING is

desirable when you’re trying to improve performance of large data loads.

What happens if you have a media failure after you’ve populated a table in NOLOGGING mode (and

before you’ve made a backup of the table)? After a restore and recovery operation, it will appear that the

table has been restored:

CHAPTER 1 ■ OPTIMIZING TABLE PERFORMANCE

SQL> desc f_regs;

Name Null? Type

----------------------------------------- -------- ----------------------------

REG_ID NUMBER

REG_NAME VARCHAR2(2000)

However, when executing a query that scans every block in the table, an error is thrown.

SQL> select * from f_regs;

This indicates that there is logical corruption in the datafile:

ORA-01578: ORACLE data block corrupted (file # 10, block # 198)

ORA-01110: data file 10: '/ora01/dbfile/O11R2/users201.dbf'

ORA-26040: Data block was loaded using the NOLOGGING option

As the prior output indicates, the data in the table is unrecoverable. Use NOLOGGING only in situations

where the data isn’t critical or in scenarios where you can back up the data soon after it was created.

■ Tip If you’re using RMAN to back up your database, you can report on unrecoverable datafiles via the REPORT

UNRECOVERABLE command.

There are some quirks of NOLOGGING that need some explanation. You can specify logging

characteristics at the database, tablespace, and object levels. If your database has been enabled to force

logging, then this overrides any NOLOGGING specified for a table. If you specify a logging clause at the

tablespace level, it sets the default logging for any CREATE TABLE statements that don’t explicitly use a

logging clause.

You can verify the logging mode of the database as follows:

SQL> select name, log_mode, force_logging from v$database;

The next statement verifies the logging mode of a tablespace:

SQL> select tablespace_name, logging from dba_tablespaces;

And this example verifies the logging mode of a table:

SQL> select owner, table_name, logging from dba_tables where logging = 'NO';

How do you tell whether Oracle logged redo for an operation? One way is to measure the amount of

redo generated for an operation with logging enabled vs. operating in NOLOGGING mode. If you have a

development environment for testing, you can monitor how often the redo logs switch while the

transactions are taking place. Another simple test is to measure how long the operation takes with and

without logging. The operation performed in NOLOGGING mode should occur faster because a minimal

amount of redo is generated during the load.

CHAPTER 1 ■ OPTIMIZING TABLE PERFORMANCE

1-7. Efficiently Removing Table Data

Problem

You’re experiencing performance issues when deleting data from a table. You want to remove data as

efficiently as possible.

Solution

You can use either the TRUNCATE statement or the DELETE statement to remove records from a table.

TRUNCATE is usually more efficient but has some side effects that you must be aware of. For example,

TRUNCATE is a DDL statement. This means Oracle automatically commits the statement (and the current

transaction) after it runs, so there is no way to roll back a TRUNCATE statement. Because a TRUNCATE

statement is DDL, you can’t truncate two separate tables as one transaction.

This example uses a TRUNCATE statement to remove all data from the COMPUTER_SYSTEMS table:

SQL> truncate table computer_systems;

When truncating a table, by default all space is de-allocated for the table except the space defined by

the MINEXTENTS table-storage parameter. If you don’t want the TRUNCATE statement to de-allocate the

currently allocated extents, then use the REUSE STORAGE clause:

SQL> truncate table computer_systems reuse storage;

You can query the DBA/ALL/USER_EXTENTS views to verify if the extents have been de-allocated (or

not)—for example:

select count(*)

from user_extents where segment_name = 'COMPUTER_SYSTEMS';

How It Works

If you need the option of choosing to roll back (instead of committing) when removing data, then you

should use the DELETE statement. However, the DELETE statement has the disadvantage that it generates a

great deal of undo and redo information. Thus for large tables, a TRUNCATE statement is usually the most

efficient way to remove data.

Another characteristic of the TRUNCATE statement is that it sets the high-water mark of a table back to

zero. When you use a DELETE statement to remove data from a table, the high-water mark doesn’t

change. One advantage of using a TRUNCATE statement and resetting the high-water mark is that full table

scan queries search only for rows in blocks below the high-water mark. This can have significant

performance implications for queries that perform full table scans.

CHAPTER 1 ■ OPTIMIZING TABLE PERFORMANCE

Another side effect of the TRUNCATE statement is that you can’t truncate a parent table that has a

primary key defined that is referenced by an enabled foreign-key constraint in a child table—even if the

child table contains zero rows. In this scenario, Oracle will throw this error when attempting to truncate

the parent table:

ORA-02266: unique/primary keys in table referenced by enabled foreign keys

Oracle prevents you from truncating the parent table because in a multiuser system, there is a

possibility that another session can populate the child table with rows in between the time you truncate

the child table and the time you subsequently truncate the parent table. In this situation, you must

temporarily disable the child table–referenced foreign-key constraints, issue the TRUNCATE statement,

and then re-enable the constraints.

Compare the TRUNCATE behavior to that of the DELETE statement. Oracle does allow you to use the

DELETE statement to remove rows from a parent table while the constraints are enabled that reference a

child table (assuming there are zero rows in the child table). This is because DELETE generates undo, is

read-consistent, and can be rolled back. Table 1-4 summarizes the differences between DELETE and

TRUNCATE.

If you need to use a DELETE statement, you must issue either a COMMIT or a ROLLBACK to complete the

transaction. Committing a DELETE statement makes the data changes permanent:

SQL> delete from computer_systems;

SQL> commit;

■ Note Other (sometimes not so obvious) ways of committing a transaction include issuing a subsequent DDL

statement (which implicitly commits an active transaction for a session) or normally exiting out of the client tool

(such as SQL*Plus).

If you issue a ROLLBACK statement instead of COMMIT, the table contains data as it was before the

DELETE was issued.

When working with DML statements, you can confirm the details of a transaction by querying from

the V$TRANSACTION view. For example, say that you have just inserted data into a table; before you issue a

COMMIT or ROLLBACK, you can view active transaction information for the currently connected session as

follows:

SQL> insert into computer_systems(cs_id) values(1);

SQL> select xidusn, xidsqn from v$transaction;

XIDUSN XIDSQN

---------- ----------

3 12878

SQL> commit;

SQL> select xidusn, xidsqn from v$transaction;

no rows selected

剩余584页未读，继续阅读

ccnawolf

粉丝: 4
资源: 43

Oracle 11g性能调优实战指南：快速解决数据库问题

Apress.Oracle.Database.11g.Performance.Tuning.Recipes

Oracle Database 11g Performance Tuning Recipes 英文版

Oracle.Database.10g.Performance.Tuning.Tips.and.Techniques.

oracle19c官方文档

linux performance tuning pdf

oracle performance tuning guide

oracle1017

oracle IO waits

ERROR: Cannot install tuning==2020.8.28.15.36.59 and tuning==2020.8.28.15.39.17 because these package versions have conflicting dependencies.

最新资源