Search This Blog

Saturday, September 18, 2010

Oracle update performance speed best practices

Oracle update performance speed best practices

The SQL standard for DML UPDATE statements can be complex and convoluted and there are best practices that can help you to write efficient UPDATE statements.

    * Run updates in batch mode
    * Use CTAS in lieu of large updates
    * Include the SET condition in the WHERE clause
    * Simplify the WHERE predicates
    * Have a small, separate data cache for high DML tables

Running updates in batch mode

If you can take your application offline during the monthly update, you use many update performance features to speed up the job performance:

    * Drop indexes/constraints and rebuild after mass update:  Dropping indexes before a mass update and rebuilding them afterwards can improve update performance significantly. Oracle removes index entries without re-balancing the index tree (a "logical delete"), but this is still time-consuming, especially if you have lots of indexes on the target table.  Also note that you can rebuild the dropped indexes nologging mode.

    * Parallelize the updates:  If you have an SMP server you can run Oracle parallel DML.    You can also manually parallelize the update by breaking them into multiple jobs and submit them simultaneously using dbms_scheduler, cron  or nohup.

Use CTAS in lieu of large updates

When you are updating the majority of rows in a table, using Create Table As Select (CTAS) is often more efficient performance than a standard update.  For example, assume that the following update changed 75% of the table rows:

    update
       mytab
    set
       status = 'new'
    where
       status = 'old;

In this case, a parallelized CTAS may perform far faster (Note: Make sure that you have an SMP server before using the parallel degree option):

    create table new_mytab NOLOGGING as
    select  /*+ full parallel(mytab,35)*/
       decode (status,'new','old',status,
       col2, col3, col4
    from mytab;

    -- rebuild indexes, triggers and constraints to new_mytab

    rename mytab to bkup_mytab;
    rename new_mytab to mytab;
    

Include the SET condition in the WHERE clause

This note shows a case where the developer forgot to include the SET condition in the UPDATE WHERE clause, causing high redo waits (log file parallel write waits, log file sync waits).  Simply including the existing state of the SET clause can result in a huger performance improvement for UPDATE statements:

-- zillion row update
UPDATE HISTORY SET FLAG=0 WHERE CLASS='X'

-- hundred row update
UPDATE HISTORY SET FLAG=0 WHERE CLASS='X' AND FLAG!=0
The select for update is not a good locking strategy because there are many things that can go wrong.  Instead of select for update, savvy Oracle developers will adopt alternatives mechanisms like a re-read upon update commit where the transaction re-read the rows and ensure that it has not changed since the original read.
Have a small, separate data cache for high DML tables

As more people adopt 64-bit servers with giant data buffers, we see a delay caused by the database writer process having to scan through giant data buffers seeking dirty blocks.  Many shops are replacing their platter-style disks with solid-state disks, and creating a very small data buffer, just for the updates.  The book Oracle Tuning: The Definitive Reference notes that if you are still using traditional disks, many shops segregate high-updates objects (tables & indexes) into a separate blocksize so that they can have a separate, small data buffer.

    "By segregating high activity tables into a separate, smaller data buffer, Oracle has far less RAM frames to scan for dirty block, improving the throughput and also reducing CPU consumption. This is especially important for high update tables with more than 100 row changes per second."

MOSC Note:223299.1 also embraces the importance of multiple blocksizes, listing the multiple buffer regions as among the most important tuning parameters in Oracle9i.
Simplify the WHERE clause predicates

The most common issue with updates is the requirement to have a complex SELECT statement is the where clause to identify the rows to be updated.  The best techniques for simplifying UPDATE where clauses include:

    * Predicate pushing
    * Rewriting subqueries as outer joins (if possible)
    * Oracle SQL subquery unnesting
    * Partitioning Pruning
    * Avoid IN and NOT IN - It is a good idea to discourage the use of the NOT IN clause (which invokes a sub-query) and to prefer NOT EXISTS (which invokes a correlated sub-query).

Monday, September 6, 2010

11g R2 New Features

Application Development

Application Express:
  • Application Date Format
  • Custom Themes
  • BLOB Support
  • JavaScript Libraries
  • Report Printing (Enhanced)
  • Forms Conversion
  • Security (Enhanced)
  • Runtime-Only Install
General Features:
  •  Pro*C
  •  “WITH HOLD” Option for Cursors
  •  8-byte Numeric host Variable
  •  Pro*COBOL
  •  8-byte Numeric host Variable
JDBC
  • TZ Patching
  • SecureFile Zero-Copy
OCI
  • 8-byte Numeric host Variable
Availability


  • Backup and Recovery:
  • Automatic Block Repair
  • OSB Cloud Module (Amazon Service)
  • DUPLICATE
  • Without Target Connection
  • NOREDO and “UNDO TABLSPACE”
  • Tablespace Checks
  • TSPITR (Enhanced)
  • “SET NEWNAME” format and clauses  
Online Maintenance and Upgrade:
  • Edition Based Redefinition
  • “CREATE OR REPLACE TYPE” to Allow FORCE
  • Fine-Grained Dependencies for Triggers
  • IGNORE_ROW_ON_DUPKEY_INDEX Hint (Insert Statement)

 Data Guard:

  • Compressed Table (Logical Standby and LogMiner)
  • Real-Time Query Lagging
  • FAN Support
  • 30 Standby Databases
BI and Data Warehousing

Analytics:
  • Analytic Functions 2.0
  • Recursive WITH
Data Loading:
  • EXECUTE Privilege for Directory Objects
  • ORACLE_LOADER Access Driver Extended
Partitioning:
  • Virtual Columns in PK or FK
  • System-Managed Indexes for List Partitions
Performance and Scalability:
  • In-Memory Parallel Execution
  • Auto POD
  • DBMS_PARALLEL_EXECUTE Package
  • On-Commit Fast Refresh (Significant Performance Improvement)
Warehouse Builder:
  • Mapping Editor
  • Advanced Find
  • Copy and Paste of Operators and Attributes
  • Grouping and Spotlighting of Objects
  • Business Intelligence Tool Integration (BI Standard and Enterprise)
  • Current Configuration Dropdown in Design Center
  • Flat File Imports (Enhanced Support)
  • Table Function (Enhanced Support)
  • Experts Available in Editor Menu
  • Expression Editing in Operator Edit
  • Management/UI of Locations Registered (Enhanced)
  • Key Lookup Operator (Enhanced)
  • Mapping Debugger (Enhanced)
  • JDeveloper-Style UI
  • Operator References Included in Generated PL/SQL Code
  • Quick Mapper
  • Repository Browser (Enhanced)
  • Simplified Repository Upgrades
  • LONG Data Type Data Extraction (Supported)
  • Sub-Queries in Join Operator (Supported)
Clustering :RAC in-use
  • Cluster Time Service
  • Oracle Cluster Registry (Enhancements)
  • Voting Disk (Enhancements)
  • Grid Plug and Play
  • Oracle Restart
  • Policy-Based Cluster/Capacity Management
  • Clusterware Resource Modeling (Enhanced)
  • Role-Separated Management
  • Agent Development Framework
  • Zero Downtime Patching (Clusterware and RAC)
  • Out-of-Place Clusterware Upgrade
  • Server Control (SRVCTL)
  • Oracle Restart (Support)
  • Grid Plug and Play (Support)
  • Single Instance Database in a Cluster (Support)
  • Universal Connection Pool (UCP)
  • Data Guard Integration
  • RAC Integration
  • JDBC (Support)
  • Enterprise Manager
  • Clusterware Resource Management
  • Provisioning (Clusterware and RAC)
  • Grid Plug and Play (Support)
  • Oracle Restart (Support)
  • Universal Installer
  • Removing RAC (Support)
  • Installation Rollback
  • Oracle Restart Integration
  • Out-of-Place Upgrade (Support)
  • Fixup Scripts and Grid Infrastructure Checks
  • Typical Installation (Option)
  • Single Client Access Name (SCAN)
  • Downgrade Database Configured with DBControl
  • Java API for RAC FAN Events
  • Intelligent Platform Management Interface (IPMI) Integration
  • ASM Integration with Clusterware
  • Configuration Assistants
  • New Features (Support)
  • Removing RAC (Support)
  • Cluster Verification Utility
  • New Features (Enhanced)
  • Integration into Universal Installer

Database General:
  • Flashback Data Archive DDL (Support)
  • Instance Caging
  • Scheduler:
  • E-mail Notification
  • File Watcher
  • Multiple Destination Jobs
  • Remote Database Jobs
  • Utilities:
  • Data Pump Legacy Mode
  • IPv6:
  • JDBC Thin Clients (Support)
  • JVM and Java Debuggers (Support)
  • Smart Flash Cache
  • Stored Outline Migration to SQL Plan Management
  • Client Result Cache Using Table Annotations (Support)
  • 4 KB Sector Disk Drive (Support)
Security
  • Audit Data Management:
  • Audit Trail Cleanup
  • Encryption Key Management
  • Tablespace Master Key Rekey
Server Management
  • Automatic Storage Management (ASM) for all Data
  • ASM Dynamic Volume Manager (DVM)
  • ASM Cluster File System (ACFS)
  • ASM FS Snapshot
  • Oracle Cluster Registry (OCR) and Voting Disk
  • ASM Intelligent Data Placement
  • ASM Configuration Assistant
  • ASM File Access Control
  • ASMCMD (Extension)
  • Enterprise Manager Integration
  • ACFS (Support)
  • ASM File Access Control
  • EMCA for New RAC Configurations (Support)
  • Patch Application with DBConsole (Single-Instance)
  • Automatic Patching of Time Stamp w/ Time Zone Data
  • Prevent Data Loss with Local Time Zone Data Type
  • Segment Creation on Demand
  • Zero-Sized Unusable Indexes and Index Partitions
  • Metadata SXML Comparison Tool
  • Compare Period Report
  • Compare SQL Tuning Sets
  • Sampling for Active Data Guard
  • Exadata Simulation
  • Global RAC ASH Report and ADDM Backward Compatability
  • Replay capture on MTS and Streams (Support)
  • SQL Tuning Advisor (Support)
  • Parallel Query (Auto DOP)
  • Action Plan Recording
  • Export SQL Tuning Set (STS)
  • Replay Synchronization Controls
Multimedia and DICOM:
  • Extraction of Requested Attributes Only
  • Client –Side DICOM Attribute Extraction
  • DICOM (Enhancements)
  • Watermarking and Image Processing (Enhancements)

Spatial:
  • 3D Visualization Framework
  • Network Data Model (Enhancements)
  • GeoRaster JAVA API
  • Raster Reprojections and Ground Control Point-Based Georeferencing
  • Routing and Geocoding (Enhancements)

SecureFiles:

  • Database File System (DBFS)
  • LZO Compression (Support)

XML DB:

  • Binary XML (Enhancements)
  • XML DB Repository Performance (Enhancements)
  • XMLIndex (Enhancements)
  • XMLType Partitioning











Oracle JDBC encryption

How to configure encrypted connection to encrypt data over the wire, and check summing, which prevents certain classes of network attacks. .

Preliminary - Gather Information

DBA must have installed the Oracle server software with the Advanced Security option, and the DBA must have configured the database listener to at least allow encrypted sessions and check summing.

Install Oracle Client

The standard Oracle client software is sufficient on the client side. There are no special client editions of the software to consider. Install the client software and test it as per usual.


Configuring the Client for Encryption

Edit your $ORACLE_HOME/network/admin/sqlnet.ora file on the client machine. For this example, we'll assume DBA has setup your instance to use RC4_256 for encryption and MD5 for checksums. We'll also assume you want your connection to be encrypted, or else let the connection attempt fail.

To configure this, add the following lines to your sqlnet.ora :

SQLNET.CRYPTO_CHECKSUM_TYPES_CLIENT = (MD5)
SQLNET.ENCRYPTION_TYPES_CLIENT = (RC4_256)
SQLNET.ENCRYPTION_CLIENT = required
SQLNET.CRYPTO_CHECKSUM_CLIENT = required
SQLNET.CRYPTO_SEED = 'insert a random string from 10-70 characters here'

The full set of options for both ENCRYPTION_CLIENT and CRYPTO_CHECKSUM_CLIENT are: accepted, rejected, requested, required

The full set of options for ENCRYPTION_TYPES_CLIENT are: RC4_256, RC4_128, RC4_56, RC4_40, AES256, AES192, AES128, 3DES168, 3DES112, DES, DES40 (Note: Oracle recommends RC4_256 for performance reasons.)

The full set of options for CRYPTO_CHECKSUM_TYPES_CLIENT are: MD5, SHA1

The CRYPTO_SEED must be quoted in single or double quotes.


It's possible to specify more than one encryption or checksum algorithm. To do so, separate them by commas like so:

SQLNET.ENCRYPTION_TYPES_CLIENT = (AES256,RC4_256,DES)

They will be tried in order until a match is found. If no match is found, then the connection is refused (if either client or server require encryption) or allowed but is unencrypted (all other cases). This may be useful if two different instances have a disjoint set of accepted encryption algorithms.

Note: There is no plausible reason we can think of to configure the client or server to always reject encrypted connections. However, if one side is set to reject, and the other to require, the result is what you'd expect: a connection can never be established.

Verifying encryption is working

To verify the encryption is working, it's both tedious and error prone to use a packet sniffer to verify the connect to the Oracle instance is encrypted. It is not trivial to distinguish compressed data streams from encrypted ones.

A more straightforward and reasonably solid approach is to enable client-side tracing and let Oracle's diagnostic information verify if encryption is working. To do this, edit your sqlnet.ora file, and temporarily add the following lines:

TRACE_LEVEL_CLIENT=10
TRACE_DIRECTORY_CLIENT=(a fully qualified directory to which you have write permissions)

NOTE: On Windows, TRACE_DIRECTORY_CLIENT is reported to be ignored. Instead, look under:

"C:\Documents and Settings\\Oracle\oradiag_\diag\clients\user_\host_<...>\trace\"

Use sqlplus to login to your instance. We recommend not using more elaborate or GUI tools for this test, as many such tools tend to automatically issue a number of queries on login, and will make the trace logs extremely verbose. After you have logged in successfully, logout.

In the TRACE_DIRECTORY_CLIENT, you should see a file named cli_NNNNN.trc, where NNNNN is an integer. (Note for Windows: the trace files are reported to begin with ora_ rather than cli_.) Grep or visually inspect this file, looking for the string 'encryption'. For example, a session that has successfully set up an encrypted connection will contain a line such as:

[24-OCT-2006 10:00:32:087] na_tns: encryption is active, using RC4_256

A cleartext connection would instead look like:

[24-OCT-2006 10:07:34:707] na_tns: encryption is not active

If you are familiar with Oracle client traces, please note encrypted session negotiation will not turn up at trace levels below 10. You must use an ADMIN or SUPPORT trace level.

Once testing is complete, you will probably want to remove TRACE_LEVEL_CLIENT and TRACE_DIRECTORY_CLIENT from your sqlnet.ora file.

Reference:
http://download.oracle.com/docs/cd/B28359_01/java.111/b31224/clntsec.htm#insertedID5