4 Optimizing PL/SQL Performance

This chapter discusses the techniques for improving PL/SQL performance in Oracle HTTP Server.

This chapter contains the following sections:

• PL/SQL Performance in Oracle HTTP Server - Overview
• Process-Based and Thread-Based Operation in Oracle HTTP Server
• Performance Tuning Issues for mod_plsql
• Performance Tuning Areas in mod_plsql
• Using Caching with PL/SQL Based Web Applications
• Tuning File System Cache to Improve Caching Performance
• Oracle HTTP Server Directives

4.1 PL/SQL Performance in Oracle HTTP Server - Overview

This chapter describes several techniques to improve the performance of PL/SQL based Web applications in Oracle HTTP Server.

Table 4-1 lists recommendations for Database Access Descriptor (DAD) parameters and settings. By default, these DAD parameters are specified in the file dads.conf. On UNIX systems, this is in the ORACLE_HOME/Apache/modplsql/conf directory. On Windows systems, by default, this file is in the directory ORACLE_HOME\Apache\Apache\modplsql\conf directory. The file dads.README in this directory describes the DAD parameters in detail.

Table 4-1 Database Access Descriptor (DAD) Parameters Recommended Setting Summary

Parameter	Recommended Setting
PlsqlAlwaysDescribeProcedure	Set this to off for best performance. Default Value: off
PlsqlDatabaseConnectString	For newer DADs, use the ServiceNameFormat. Use the SIDFormat only for backward compatibility. Note: for HA configurations of the database, it is recommended that the connect string parameter gets resolved through an LDAP lookup.
PlsqlFetchBufferSize	For multibyte character sets like Japanese or Chinese, setting this to 256 should provide better performance Default Value: 128
PlsqlIdleSessionCleanupInterval	Increasing this parameter allows pooled database connections to remain available, in the pool, for the specified time Default Value: 15 (minutes) See Also: Section 4.3.3, "Restarting the mod_plsql Connection Pool"
PlsqlLogEnable	This parameter should be set to Off unless recommended by Oracle support for debugging purposes Default Value: off
PlsqlMaxRequestsPerSession	If the PL/SQL based Web application does not leak resources or memory, this parameter can be set to a higher value (for example, 5000). Default Value: 1000 See Also: Section 4.3.2, "Closing Pooled Database Sessions" and Section 4.4.2, "Connection Pooling Tips and Oracle HTTP Server Configuration".
PlsqlNLSLanguage	Set this parameter to match the database Globalization Support parameters to eliminate overheads in character set conversions in Oracle Net Services
PlsqlSessionStateManagement	Set this parameter to StatelessWithFastResetPackageState if the database is 8.1.7.2 or later.

Table 4-2 lists mod_plsql caching options and the sections that describe these caching options.

Table 4-2 Caching Options

Option	Description
Expires Technique	Best performance - for content that changes predictably. See Also: Section 4.5.2, "Using the Expires Technique"
Validation technique	Good performance - for content that changes unpredictably. See Also: Section 4.5.1, "Using the Validation Technique"
System-level caching	Improves performance by caching one copy for everyone on system. See Also: Section 4.5.3, "System- and User-level Caching with PL/SQL Based Web Applications"

4.2 Process-Based and Thread-Based Operation in Oracle HTTP Server

This chapter describes PL/SQL performance issues that apply on platforms where the Oracle HTTP Server is process-based and thread-based. On a process-based Oracle HTTP Server, such as those running on UNIX-based platforms, each process servers all types of HTTP requests, including servlets and PL/SQL, static files. On a thread-based Oracle HTTP Server, such as Windows-based platforms, there is just one Oracle HTTP Server process with multiple threads within the process; individual threads can be used serve all types of HTTP requests.

4.3 Performance Tuning Issues for mod_plsql

When tuning mod_plsql to improve the performance of PL/SQL based Web applications, it is important to be familiar with some mod_plsql internals. This section presents a basic overview of some mod_plsql functionality.

This section covers the following topics:

• Connection Pooling with mod_plsql
• Closing Pooled Database Sessions
• Restarting the mod_plsql Connection Pool

4.3.1 Connection Pooling with mod_plsql

The Database Server connection pooling logic supplied with mod_plsql can be best explained with an example.

For example, consider the following typical scenario:

1. The Oracle HTTP Server listener is started. There are no database connections in the connection pool maintained by mod_plsql.
2. A browser makes a mod_plsql request (R1) for Database Access Descriptor (DAD) D1.
3. One of the Oracle HTTP Server processes (httpd process P1) starts servicing the request R1.
4. mod_plsql in process P1 checks its connection pool and finds that there are no database connections in its pool for that user request.
5. Based on the information in DAD D1, mod_plsql in process P1 opens a new database connection, services the PL/SQL request, and adds the database connection to its pool.
6. From this point on, all subsequent requests to process P1 for DAD D1 can now make use of the database connection pooled by mod_plsql.
7. If a request for DAD D1 gets picked up by another process (process P2), then mod_plsql in process P2 opens its own database connection, services the request, and adds the database connection to its pool.
8. From this point on, all subsequent requests to process P2 for DAD D1 can now make use of the database connection pooled by mod_plsql.
9. Now, assume that a request R2 is made for DAD D2 and this request gets routed to process P1.
10. mod_plsql in process P1 does not have any database connections pooled for DAD D2, and a new database session is created for DAD D2 and pooled after servicing the request. Process P1 now has two database connections pooled, one for DAD D1 and another for DAD D2.

    The important details in the example shown in steps 1-10 are:

    1. Each Oracle HTTP Server process serves all types of requests, such as static files requests, servlet requests, and mod_plsql requests. There is no control on which Oracle HTTP Server process services the next request.
    2. One Oracle HTTP Server process cannot use or share the connection pool created by another process.

       Note: On Windows systems, the Oracle HTTP Server is just one process. Therefore, the connection pool is shared and can be used across threads.

    3. Each Oracle HTTP Server process pools at most one database connection for each DAD.
    4. User sessions are switched within a pooled database connection for a DAD. For DADs based on Oracle Application Server Single Sign-On (SSO), proxy authentication is used to switch the user session. For non-SSO users, using HTTP basic authentication with the username and password not in the DAD, users are re-authenticated on the same connection.
    5. Multiple DADs may point to the same database instance, but database connections are not shared across DADs even within the same process.
    6. Unused DADs do not result in any database connections.

In the worst-case scenario, the total number of database connections pooled by mod_plsql is a factor of the total number of active DADs multiplied by the number of Oracle HTTP Server (httpd) processes running at any given time for a single Oracle HTTP Server instance. If you have configured the Oracle HTTP Server processes to a high number, you need to configure the back-end database to handle a corresponding number of database sessions, and remember that this configuration value needs to be multiplied times the number of Oracle HTTP Server instances that use the back-end database.

For example, if there are three Oracle HTTP Server instances configured to spawn a maximum of 50 httpd processes each, plus two active DADs, you need to set up the database to allow 300 (3*50*2) sessions. This number does not include any sessions that are needed to allow other Web applications to connect.

On UNIX systems, database connections cannot be shared across httpd processes, and process-based platforms have more of a Connection Reuse feature than Connection Pooling. Note that this is an artifact of the process-model in Oracle HTTP Server.

4.3.2 Closing Pooled Database Sessions

Pooled database sessions are closed under the following circumstances:

1. When a pooled connection has been used to serve a configured number of requests.

   By default each connection pooled by mod_plsql is used to service a maximum of 1000 requests and then the database connection is shut down and reestablished. This is done to make sure that any resource leaks in the PL/SQL based Web application, or in the Oracle client server side, do not adversely affect the system. To change the default value of 1000 by tuning the DAD configuration parameter PlsqlMaxRequestsPerSession.

2. When a pooled connection has been idle for an extended period of time.

   By default, each pooled connection gets automatically cleaned up after 15 minutes of idle time. This operation is performed by the cleanup thread in mod_plsql. For heavily loaded sites, each connection could be used at least once every 15 minutes and the connection cleanup might not happen for a long period of time. In such a case, the connection would be cleaned up based on the configuration value of PlsqlMaxRequestsPerSession. Change the default value of 15 minutes by tuning the mod_plsql configuration parameter PlsqlIdleSessionCleanupInterval. Consider increasing the default for better performance in cases where the site is not heavily loaded.

3. On UNIX systems, when the Oracle HTTP Server process goes down.

   On UNIX systems, the Oracle HTTP Server configuration parameter MaxRequestsPerChild governs when an Oracle HTTP Server process will be shut down. For example, if this parameter is set to 5000, each Oracle HTTP Server process would serve exactly 5000 requests before it is shut down. Oracle HTTP Server processes could also start up and shut down as part of Oracle HTTP Server maintenance based on the configuration parameters MinSpareServers, MaxSpareServers, and MaxClients. For mod_plsql connection pooling to be effective, it is extremely important that Oracle HTTP Server is configured so that each Oracle HTTP Server process remains active for some period of time. An incorrect configuration of Oracle HTTP Server could result in a setup where Oracle HTTP Server processes are heavily started up and shut down. Such a configuration would require that each new Oracle HTTP Server process replenish the connection pool before subsequent requests gain any benefit of pooling.

   See Also: Chapter 6 "Oracle HTTP Server Modules" in the Oracle HTTP Server Administrator's Guide.

4.3.3 Restarting the mod_plsql Connection Pool

This depends primarily on the amount of time the database is shut down. If the database is restarted after more than 15 minutes from being shut down, the users do not experience any problems when trying to use the Oracle HTTP Server listener. This is because the cleanup thread in mod_plsql cleans up database sessions that are unused for more than 15 minutes. The time specified for cleaning up idle sessions is tunable using the PlsqlIdleSessionCleanupInterval, configuration parameter (the default value is 15 minutes).

If the database is restarted in less than 15 minutes, then a few initial requests return with errors, but the system quickly becomes usable again. The number of requests that experience failure is equal to the number of connections that were pooled by mod_plsql.

4.4 Performance Tuning Areas in mod_plsql

While using mod_plsql, there are three areas that affect performance and scalability:

• PL/SQL Based Web Application Development Considerations
• Connection Pooling Tips and Oracle HTTP Server Configuration
• Tuning the Number of Database Sessions

4.4.1 PL/SQL Based Web Application Development Considerations

PL/SQL Gateway users should consider the following topics when developing PL/SQL based Web applications:

1. Manage the use of Database Access Descriptors (DADs)

   Try to restrict the number of DADs that each Oracle HTTP Server node uses.

   Note: Performance is not affected if there are DADs that are not being used.

2. Use of Nested Tables

   PL/SQL provides the ability to create tables. To build PL/SQL tables, you build a table that gives the datatype of the table, as well as the index of the table. The index of the table is the binary integer ranging from -2147483647 to +2147483647. This table index option is known as sparsity, and allows meaningful index numbers such as customer numbers, employee number, or other useful index keys. Use PL/SQL tables to process large amounts of data.

   PL/SQL provides TABLE and VARRAY (variable size array) collection types. The TABLE collection type is called a nested table. Nested tables are unlimited in size and can be sparse, which means that elements within the nested table can be deleted using the DELETE procedure. Variable size arrays have a maximum size and maintain their order and subscript when stored in the database. Nested table data is stored in a system table that is associated with the nested table. Variable size arrays are suited for batch operations in which the application processes the data in batch array style. Nested tables make for efficient queries by storing the nested table in a storage table, where each element maps to a row in the storage table.

3. Use procedure naming overloading with caution

   PL/SQL based Web applications should use the procedure name overloading feature with caution. It is best if procedure name overloading is avoided by having multiple procedures with different names.

4. Consider rewriting applications where there is significant overhead in determining the type parameters

   PL/SQL based Web applications should be aware of the overhead in trying to execute procedures where the URL does not provide enough details to know about the type of the parameter, such as scalar or array. In such cases, the first attempt to execute a procedure fails and the procedure signature needs to be described before it can be executed by mod_plsql.

5. Use procedures with 2-parameter style flexible parameter passing

   Procedures should make use of the more performant 2-parameter style flexible parameter passing rather than the 4-parameter style parameter passing

   See Also: Oracle Database Application Developer's Guide - Fundamentals

4.4.2 Connection Pooling Tips and Oracle HTTP Server Configuration

Consider the following topics when configuring connection pooling with Oracle HTTP Server:

1. Using the default connections pooling and setting values for PlsqlMaxRequestsPerSession

   Creating new database connections is an expensive operation and it is best if every request does not have to open and close it own database connections. The optimal technique is to make sure that database connections opened in one request are reused in subsequent requests. In some rare situations, where a database is accessed very infrequently and performance is not a major concern, connection pooling can be disabled. For example, if the administrator accesses a site infrequently to perform some administration tasks, then the DAD used to access the administrator applications can choose to disable connection pooling. To disable connection pooling, set the DAD parameter PlsqlMaxRequestsPerSession to the value 1.

   Note: Setting PlsqlMaxRequestsPerSession to the value 1 reduces the number of available database sessions and may impact performance.

2. On UNIX systems, Oracle HTTP Server configuration should be properly tuned so that once processes are started up, the processes remain up for a while. Otherwise, the connection pooling in mod_plsql is rendered useless. The Oracle HTTP Server listener should not have to continually start up and shut down processes. A proper load analysis should be performed of the site to determine what the average load on the Web site. The Oracle HTTP Server configuration should be tuned such that the number of httpd processes can handle the average load on the system. In addition, the configuration parameter MaxClients in the httpd.conf file should be able to handle random load spikes as well.
3. On UNIX systems, Oracle HTTP Server processes should be configured so that processes are eventually killed and restarted. This is required to manage any possible memory leaks in various components accessed through the Oracle HTTP Server. This is specifically required in mod_plsql to ensure that any database session resource leaks do not cause a problem. Make sure that MaxRequestsPerChild configuration parameter is set to a high number. For PL/SQL based Web applications, this should not be set to 0.

   Note: On Windows systems, to assure that database session resource leaks do not cause problems, the Oracle HTTP Server needs to be periodically restarted.

4. For heavily loaded sites, the Oracle HTTP Server configuration parameter KeepAlive should be disabled. This ensures that each process is available to service requests from other clients as soon as a process is done with servicing the current request. For sites which are not heavily loaded, and where it is guaranteed that the number of Oracle HTTP Server processes are always greater than the number of simultaneous requests to the Oracle HTTP Server listener, enabling the KeepAlive parameter results in performance improvements. In such cases, make sure to tune the KeepAliveTimeout parameter appropriately.
5. You may want to lower the value of Timeout in the Oracle HTTP Server configuration. This ensures that Oracle HTTP Server processes are freed up earlier if a client is not responding in a timely manner. Do not set this value too low, otherwise slower responding clients could time out.
6. Most Web sites have many static image files, which are displayed in each screen for a consistent user interface. Such files rarely change and you can reduce a considerable load on the system by tagging each image served by the Oracle HTTP Server listener with mod_expires. You should also consider front-ending your Web site with Oracle Application Server Web Cache.
   • How do I know if the Web site can benefit from the use of mod_expires?
     • Use Netscape, or any browser that enables you to view page caching information, and visit several heavily accessed Web pages on the site. On each page, right click the mouse and select View Info from the pop up menu (or the equivalent command for your browser). If the top panel in the page information window lists many different images and static content, then the site could benefit from the use of mod_expires.
     • You can also check the Oracle HTTP Server access logs to see what percentage of requests result in HTTP 304 (Not Modified) status. Use the grep utility to search for 304 in the access_log and divide this resulting number of lines by the total number of lines in the access_log. If this percentage is high, then the site could benefit from the use of mod_expires.
   • How do I tag static files with the Expires header?
     • Locate the Location directive used to serve your static image files. Add the ExpiresActive and ExpiresDefault directives to it.

       Alias /images/ "/u01/app/oracle/myimages/" 
       <Directory "/u01/app/oracle/myimages/"> 
            AllowOverride None 
            Order allow, deny 
            Allow from all 
            ExpiresActive On 
            ExpiresDefault A2592000
       </Directory>
       
       

       The browser caches all static files served off the /images path for 30 days from now. Refer to the Oracle HTTP Server Administrator's Guide for more details.

   • How do I know if the static files are being tagged with the Expires header?
     • Using Netscape, or the browser of your choice, clean up all the cached files in the browser.
     • Visit a Web page that should have images tagged with the Expires header. Right click the mouse on the page and select View Info, from the pop up menu. or use the equivalent command for your browser.
     • In the top panel of the page information, select an image that should be tagged with the Expires header.
     • Review the information displayed in the bottom panel. The Expires header should be set to a valid date. If this entry is No date given, then the file is not being tagged with the Expires header.

4.4.3 Tuning the Number of Database Sessions

Consider the following topics when tuning the number of database sessions:

1. The processes and sessions parameters in the Oracle init$SID.ora configuration file should be set so that Oracle is able to handle the maximum number of database sessions. This number should be proportional to the number of DADs times the maximum number of Oracle HTTP Server processes, times the number of Oracle HTTP Server instances.
2. Using a two-listener strategy or using a shared server reduces the number of database sessions. See Section 4.4.4, "Two-Listener Strategy".
3. On UNIX platforms, the connection pool is not shared across Oracle HTTP Server processes. For this reason, it is recommended that the application use as few DADs as possible.

4.4.4 Two-Listener Strategy

On platforms where the Oracle HTTP Server is process-based, such as all UNIX-based platforms, each process serves all types of HTTP requests, including servlets, PLSQL, static files, and CGI. In a single Oracle HTTP Server listener setup, each httpd process maintains its own connection pool to the database. The maximum number of database sessions is governed by the setting in httpd.conf configuration file for StartServers, MinSpareServers, and MaxSpareServers, plus the load on the system. This architecture does not allow for tuning the number of database sessions based on the number of mod_plsql requests. To tune the number of database sessions based on the number of mod_plsql requests, install a separate HTTP listener for mod_plsql requests only. This approach greatly reduces the number of database sessions that are needed to serve mod_plsql requests.

For example, assume a main Oracle HTTP Server listener is running on port 7777 of mylsnr1.mycompany.com. First, you can install another Oracle HTTP Server listener on port 8888 on mylsnr2.mycompany.com. Next, redirect all mod_plsql requests made to mylsnr1.mycompany.com:7777 to the second listener on mylsnr2.mycompany.com:8888. Review the following steps:

1. To redirect all PL/SQL requests for mylsnr1.mycompany.com:7777 to mylsnr2.mycompany.com:8888, make the following configuration changes:
   1. For the Oracle HTTP Server listener running on Port 7777, edit ORACLE_HOME/Apache/modplsql/conf/plsql.conf file. Comment out the following line by putting a # in front of the line:

      #LoadModule plsql_module...
      
      

   2. Copy the DAD location used to service PL/SQL requests in mylsnr1.mycompany.com to the configuration file ORACLE_HOME/Apache/modplsql/conf/dads.conf in mylsnr2.mycompany.com.

      Comment out the DAD location configuration parameters on mylsnr1.mycompany.com by prefixing the line with a "#" character.

      #<Location /pls/dad> 
      #... 
      #</Location>
      
      

   3. Configure this listener to forward all mod_plsql requests for this DAD location to the second listener by adding the following line in dads.conf:

      ProxyPass /pls/dad http://mylsnr2.mycompany.com:8888/pls/dad
      
      

   Repeat the configuration procedures for all DAD Locations.

2. Because the PL/SQL procedures generate URLs that are displayed in the browser, it is important that all URLs are constructed without any references to the internal mod_plsql listener on mylsnr2.mycompany.com:8888. Depending on how the URLs are being generated in the PL/SQL based Web application, there are three options:
   • If the URLs are hard-coded into the application, make sure that they are always generated using the hard-coded values as HOST=mylsnr1.mycompany.com and PORT=7777. No change would be required for this scenario.
   • If the PL/SQL based Web applications always use the CGI environment variables SERVER_NAME and SERVER_PORT, then it is easy to change the configuration of the listener on mylsnr2.mycompany.com. Edit the file and change the lines ServerName and Port in the ORACLE_HOME/Apache/Apache/conf/httpd.conf file for the second listener as follows:

     ServerName mylsnr1.mycompany.com (was mylsnr2.mycompany.com)
     Port 7777                        (was 8888)
     
     

   • If the URLs are being generated using the CGI environment variable HTTP_HOST, you need to override the CGI environment variables for the Oracle HTTP Server listener running on Port 8888. Add the following lines to the ORACLE_HOME/Apache/modplsql/conf/dads.conf file for each DAD to override the default CGI environment variables HOST, SERVER_NAME, and SERVER_PORT:

     PlsqlCGIEnvironmentList  SERVER_NAME  mylsnr1.mycompany.com 
     PlsqlCGIEnvironmentList  SERVER_PORT  7777 
     PlsqlCGIEnvironmentList  HOST mylsnr1.us.oracle.com:7777
     
     

   In all cases, the intent is to fool the application to generate URLs as if there never was a second listener.

3. Test the setup and make sure that you can access all the DADs without any problems.
4. In this setup, the main listener mylsnr1.mycompany.com can be configured based on the total load on the Oracle HTTP Server listener. The second listener on mylsnr2.mycompany.com can be fine-tuned based on just the mod_plsql requests being made.

4.4.5 Overhead Problems

While executing some of the stored procedures, mod_plsql may incur a Describe overhead, which would result in two extra round trips to the database for a successful execution. This has performance implications.

4.4.5.1 The Describe Overhead

In order to execute stored procedures, mod_plsql needs to know about the datatype of the parameters being passed in. Based on this information, mod_plsql binds each parameter either as an array or as a scalar. One way to know the procedure signature is to describe the procedure before executing it. However, this approach is not efficient because every procedure has to be described before execution. To avoid the describe overhead, mod_plsql looks at the number of parameters passed for each parameter name. It uses this information to assume the datatype of each variable. The logic is simply that if there is a single value being passed, then the parameter is a scalar, otherwise it is an array. This works for most cases but fails if there is an attempt to pass a single value for an array parameter or pass multiple values for a scalar. In such cases, the first attempt to execute the PL/SQL procedure fails. mod_plsql issues a Describe call to get the signature of the PL/SQL procedure and binds each parameter based on the information retrieved from the Describe operation. The procedure is re-executed and results are sent back.

This Describe call occurs transparently to the procedure, but internally mod_plsql has encountered two extra round trips, one for the failed execute call and the other for the describe call.

4.4.5.2 Avoiding the Describe Overhead

You can avoid performance problems with the following:

• Use flexible parameter passing.
• Always ensure that you pass multiple values for arrays. For single values, you can pass dummy values that are ignored by the procedure.
• Use the following workaround, which defines a two-parameter style procedure which defaults the unused variables.
  1. Define a scalar equivalent of your procedure, which internally calls the original procedure. For example, the original package could be similar to the following example:

     CREATE OR REPLACE PACKAGE testpkg AS
      TYPE myArrayType is TABLE of VARCHAR2(32767) INDEX BY binary_ integer;
      PROCEDURE arrayproc (arr myArrayType);
     END testpkg;
     /
     
     

  2. If you are making URL calls like /pls/.../testpkg.arrayproc? arr= 1, change the specification to be similar to the following:

     CREATE OR REPLACE PACKAGE testpkg AS
      TYPE myArrayType is TABLE of VARCHAR2( 32767) INDEX BY binary_integer;
       PROCEDURE arrayproc (arr varchar2);
       PROCEDURE arrayproc (arr myArrayType);
     END testpkg;
     /
     
     

  3. The procedure arrayproc should be similar to:

     CREATE OR REPLACE PACKAGE BODY testpkg AS
     PROCEDURE arrayproc (arr varchar2) IS
       localArr myArrayType;
     BEGIN
       localArr( 1) := arr;
       arrayproc (localArr);
     END arrayproc;
     
     
     

4.4.6 The Flexible Parameter Passing (four-parameter) Overhead

Round-trip overhead exists if a PL/SQL procedure is using the older style four-parameter interface. The PL/SQL Gateway first tries to execute the procedure by using the two-parameter interface. If this fails, the PL/SQL Gateway tries the four-parameter interface. This implies that all four-parameter interface procedures experience one extra round-trip for execution.

• Avoiding the flexible parameter passing overhead

  To avoid this overhead, it is recommended that you write corresponding wrappers that use the two-parameter interface and internally call the four-parameter interface procedures. Another option is to change the specification of the original procedure to default to the parameters that are not passed in the two-parameter interface. The four-parameter interface has been provided only for backward compatibility and will be deprecated in the future.

• Using flexible parameters and the exclamation mark

  The flexible parameter passing mode in Oracle HTTP Server expects the PL/SQL procedure to have the exclamation mark before the procedure name. Due to performance implications of the auto-detect method used in Oracle HTTP Server, the exclamation mark is now required for flexible parameter passing in Oracle HTTP Server. In Oracle HTTP Server, each procedure is described completely before being executed. The procedure Describe call determines the signature of the procedure and requires around-trip to the database. The PL/SQL Gateway in Oracle HTTP Server avoids this round trip by having end-users explicitly indicate the flexible parameter passing convention by adding the exclamation mark before the procedure.

4.5 Using Caching with PL/SQL Based Web Applications

Caching can improve the performance of PL/SQL based Web applications. To improve performance, you can cache Web content generated by PL/SQL procedures in the middle-tier and decrease the database workload.

This section covers the techniques used in caching, including the following:

• Using the Validation Technique - An application asks the server if the page has been modified since it was last presented.
• Using the Expires Technique - Based upon a specific time period, the PL/SQL based Web application determines the page will be cached, or should be generated again.
• System- and User-level Caching with PL/SQL Based Web Applications - This is valid whether you are using the Validation Technique or the Expires Technique. The level of caching is determined by whether a page is cached for a particular user or for every user in the system.

These techniques and levels are implemented using owa_cache packages located inside the PL/SQL Web Toolkit.

4.5.1 Using the Validation Technique

In general, the validation technique basically asks the server if the page has been modified since it was last presented. If it has not been modified, the cached page will be presented to the user. If the page has been modified, a new copy will be retrieved, presented to the user and then cached.

There are two methods which use the Validation Technique: Last-Modified method, and the Entity Tag method. The next two sections show how these techniques are used in the HTTP protocol. Although the PL/SQL Gateway does not use the HTTP protocol, many of the same principles are used.

4.5.1.1 Last-Modified

When a Web page is generated using the HTTP protocol, it contains a Last-Modified Response Header. This header indicates the date, relative to the server, of the content that was requested. Browsers save this date information along with the content. When subsequent requests are made for the URL of the Web page, the browser then:

1. Determines if it has a cached version.
2. Extracts the date information.
3. Generates the Request Header If-Modified-Since.
4. Sends the request the server.

Cache-enabled servers look for the If-Modified-Since header and compare it to the date of their content. If the two match, an HTTP Response status header such as "HTTP/1.1 304 Not Modified" is generated, and no content is streamed. After receiving this status code, the browser can reuse its cache entry because it has been validated.

If the two do not match, an HTTP Response header such as "HTTP/1.1 200 OK" is generated and the new content is streamed, along with a new Last-Modified Response header. Upon receipt of this status code, the browser must replace its cache entry with the new content and new date information.

4.5.1.2 Entity Tag Method

Another validation method provided by the HTTP protocol is the ETag (Entity Tag) Response and Request header. The value of this header is a string that is opaque to the browser. Servers generate this string based on their type of application. This is a more generic validation method than the If-Modified-Since header, which can only contain a date value.

The ETag method works very similar to the Last Modified method. Servers generate the ETag as part of the Response Header. The browser stores this opaque header value along with the content that is steamed back. When the next request for this content arrives, the browser passes the If-Match header with the opaque value that it stored to the server. Because the server generated this opaque value, it is able to determine what to send back to the browser. The rest is exactly like the Last-Modified validation method as described earlier.

4.5.1.3 Using the Validation Technique for mod_plsql

Using HTTP validation caching as a framework, the following is the Validation Model for mod_plsql.

PL/SQL based Web applications that want to control the content being served should use this type of caching. This technique offers some moderate performance gains. One example of this would be a Web application that serves dynamic content that could change at any given time. In this case, the Web application needs full control over what is being served. Validation caching always asks the Web application whether the cached content is stale or not before serving it back to the browser.

Figure 4-1 shows the use of the validation technique for mod_plsql.

1. The Oracle HTTP Server receives a PL/SQL procedure request from a client server. The Oracle HTTP Server routes the request to mod_plsql.
2. mod_plsql prepares the request.
3. mod_plsql invokes the PL/SQL procedure in the Web application. mod_plsql passes the usual Common Gateway Interface (CGI) environment variables to the Web application.
4. The PL/SQL procedure generates content to pass back. If the PL/SQL procedure decides that the generated content is cacheable, it calls the owa_cache procedure from the PL/SQL Web Toolkit to set the tag and cache level:

   owa_cache.set_cache(p_etag, p_level);
   

   Table 4-3 Validation Model Parameters

   Parameter	Description
   set_cache procedure	Sets up the headers to notify mod_plsql that the content being streamed back can be cached. Then, the mod_plsql caches the content on the local file system along with the tag and caching level information as it is streamed back to the browser.
   p_etag	The string that the procedure generates to tag the content.
   p_level	The caching level: SYSTEM for system level or USER for user level.

5. The HTML is returned to mod_plsql.
6. mod_plsql stores the cacheable content in its file system for the next request.
7. The Oracle HTTP Server sends the response to the client browser.

Figure 4-1 Validation Technique

Text description of the illustration asper016.gif

4.5.1.4 Second Request Using the Validation Technique

Using the Validation Technique for mod_plsql, a second request is made by the client browser for the same PL/SQL procedure.

Figure 4-2 shows the second request using the Validation Technique.

1. mod_plsql detects that it has a cached content for the request.
2. mod_plsql forwards the same tag and caching level information (from the first request) to the PL/SQL procedure as part of the CGI environment variables.
3. The PL/SQL procedure uses these caching CGI environment variables to check if the content has changed. It does so by calling the following owa_cache functions from the PL/SQL Web Toolkit:

   owa_cache.get_etag;
   owa_cache.get_level;
   
   

   These owa functions get the tag and caching level.

4. The Web application sends the caching information to mod_plsql.
5. Based on that information determines whether the content needs to be regenerated or can be served from the cache.
   1. If the content is still the same, the procedure calls the owa_cache.set_not_modified procedure and generates no content. This causes mod_plsql to use its cached content. The cached content is directly streamed back to the browser.
   2. If the content has changed, it generates the new content along with a new tag and caching level. mod_plsql replaces its stale cached copy with a new one and updates the tag and caching level information. The newly generated content is streamed back to the browser.

Figure 4-2 Validation Technique-Second Request

Text description of the illustration asper017.gif

4.5.2 Using the Expires Technique

In the validation model, mod_plsql always asks the PL/SQL procedure if it can serve the content from the cache. In the expires model, the procedure pre-establishes the content validity period. Therefore, mod_plsql can serve the content from its cache without asking the procedure. This further improves performance because no interaction with the database is required.

This caching technique offers the best performance. Use if your PL/SQL based Web application is not sensitive to serving stale content. One example of this is an application that generates news daily. The news can be set to be valid for 24 hours. Within the 24 hours, the cached content is served back without contacting the application. This is essentially the same as serving a file. After 24 hours, mod_plsql will again fetch new content from the application.

Assume the same scenario described for the Validation model, except the procedure uses the Expires model for caching.

Figure 4-3 shows the use of the expires technique for mod_plsql.

1. The Oracle HTTP Server receives a PL/SQL Server Page request from a client server. The Oracle HTTP Server routes the request to mod_plsql.
2. The request is forwarded by mod_plsql to the Oracle Database.
3. mod_plsql invokes the PL/SQL procedure in the application and passes the usual Common Gateway Interface (CGI) environment variables to the application.
4. The PL/SQL procedure generates content to pass back. If the PL/SQL procedure decides that the generated content is cacheable, it calls the owa_cache procedure from the PL/SQL Web Toolkit to set the validity period and cache level:

   owa_cache.set_expires(p_expires, p_level);
   

   Table 4-4 Expires Model Parameters

   Parameter	Description
   set_expires procedure	Sets up the headers to notify mod_plsql that Expires caching is being used. mod_plsql then caches the content to the file system along with the validity period and caching level information.
   p_expires	Number of minutes that the content is valid.
   p_level	Caching level.

5. The HTML is returned to mod_plsql.
6. mod_plsql stores the cacheable content in its file system for the next request.
7. The Oracle HTTP Server sends the response to the client browser.

Figure 4-3 The Expires Technique

Text description of the illustration asper018.gif

Second Request Using the Expires Technique

Using the same expires model explained earlier, a second request is made by the client browser for the same PL/SQL procedure.

Figure 4-4 shows the second request using the Expires Technique.

1. mod_plsql detects that it has a cached copy of the content that is expires-based.
2. mod_plsql checks the content's validity by taking the difference between the current time and the time this cache file was created.
   1. If this difference is within the validity period, the cached copy is still fresh and will be used without any database interaction. The cached content is directly streamed back to the browser.
   2. If the difference is not within the validity period, the cached copy is stale. mod_plsql invokes the PL/SQL procedure and generates new content. The procedure then decides whether to use expires-based caching again. If so, it also determines the validating period for this new content. The newly generated content is streamed back to the browser.

Figure 4-4 The Expires Technique-Second Request

Text description of the illustration asper019.gif

4.5.3 System- and User-level Caching with PL/SQL Based Web Applications

A PL/SQL procedure determines whether generated content is system-level content or user-level. This helps the PL/SQL Gateway cache to store less redundant files if more than one user is looking at the same content. It decides this by:

• For system-level content, the procedure passes the string SYSTEM as the caching level parameter to the owa_cache functions (set_cache for validation model or set_expires for expires model). This is for every user that shares the cache.

  By using system-level caching, you can save both space in your file system and time for all users in the system. One example of this would be a Web application that generates content that is intended for everybody using the Web application. By caching the content with the system-level setting, only one copy of the content is cached in the file system. Furthermore, every user on that system benefits since the content is served directory from the cache.

• For user-level content, it passes the string USER as the parameter for the caching level. This is for a specific user that is logged in. The stored cache is unique for that user. Only that user can use the cache. The type of user is determined by the authentication mode. Refer to the following table for the different types of users.

  Table 4-5 Type of User Determined by Authentication Mode

  Authentication Mode	Type of User
  Single Sign On (SSO)	Lightweight user
  Basic	Database user
  Custom	Remote user

For example, if no user customizes a PL/SQL based Web application, then the output can be stored in a system-level cache. There will be only one cache copy for every user on the system. User information is not used since the cache can be used by multiple users.

However, if a user customizes the application, a user-level cache is stored for that user only. All other users still use the system level cache. For a user-level cache hit, the user information is a criteria. A user-level cache always overrides a system-level cache.

PL/SQL Web Toolkit Functions (owa_cache package)

Your decision whether to use the Validation technique or the Expires technique determines which owa_cache functions to call.

The owa_cache package contains procedures to set and get special caching headers and environment variables. These allow developers to use the PL/SQL Gateway cache more easily. This package should already be installed in your database.

Table 4-6 lists the primary functions to call.

Table 4-6 Primary owa_cache Functions

owa Functions	Purpose
owa_cache.set_cache (p_etag IN varchar2, p_level IN varchar2)	Validation Model - Sets up the headers. p_etag parameter tags the generated content. p_level parameter is the caching level to use.
owa_cache.set_not_modified	Validation Model - Sets up the headers to notify mod_plsql to use the cached content. Only used when a validation -based cache hit occurs.
owa_cache.get_level	Validation Model - Gets the caching level, USER or SYSTEM. Returns null if the cache is not hit.
owa_cache.get_etag	Validation Model - Gets the tag associated with the cached content. Returns null if the cache is not hit.
owa_cache.set_expires( p_expires IN number, p_level IN varchar2)	Expires Model - Sets up the headers. p_expires parameter is the number of minutes the content is valid. p_level parameter is the caching level to use.

4.6 Tuning File System Cache to Improve Caching Performance

You can configure and use a File System Cache to improve the performance of PL/SQL based Web applications.

This section covers the following topics:

• Introducing File System Cache Tuning
• Enabling File System Cache
• Configuring File System Cache to Reside on a Faster File System
• Resizing File System Cache
• Configuring Cache Cleanup

4.6.1 Introducing File System Cache Tuning

This section covers mod_plsql related File System Cache tuning options. Cache contents are cached using Operating System supplied file system calls; the cached contents are not stored in the mod_plsql memory space. Using the mod_plsql File System Cache, the contents of the cache may be in memory when the Operating System supports, and the system is configured to use features such as memory disk (some UNIX platforms support memory disk based fast storage).

The information in this section can improve the performance of PL/SQL based Web applications when mod_plsql is configured to use the File System Cache.

Table 4-7 lists the cache related parameters that you can set for mod_plsql. Set these parameters in the cache.conf file that is available on UNIX in the directory, ORACLE_HOME/Apache/modplsql/conf, and on Windows, this is found in the directory, ORACLE_HOME\Apache\modplsql\conf.

Table 4-7 mod_plsql cache.conf Configuration Parameter Summary

Parameter	Description
PlsqlCacheCleanupTime	Sets the interval for running cache cleanup routines. Default: Everyday 23:00 (run cleanup routine daily at 11PM local time) See Also: Section 4.6.5, "Configuring Cache Cleanup"
PlsqlCacheDirectory	Defines the directory that holds the mod_plsql cache. Default: On UNIX systems, the default directory for the error log is: ORACLE_HOME/Apache/modplsql/cache On Windows systems, the default directory is: ORACLE_HOME\Apache\modplsql\cache See Also: Section 4.6.3, "Configuring File System Cache to Reside on a Faster File System"
PlsqlCacheEnable	Enables the file system cache. Default: On See Also: Section 4.6.2, "Enabling File System Cache"
PlsqlCacheMaxAge	Controls the aging, in days for the cache contents. Default: 30 (days) See Also: Section 4.6.4.2, "Setting the Days of Aging for Cache with PlsqlCacheMaxAge"
PlsqlCacheMaxSize	Sets the maximum size, in bytes, for an individual file stored in the cache. Default: 1048576 (1 Megabyte) See Also: Section 4.6.4.3, "Setting the Maximum File Size for a Cache File with PlsqlCacheMaxSize"
PlsqlCacheTotalSize	Limits the total size of the cache. The value is specified in bytes. Default: 20971520 (20 Megabytes) See Also: Section 4.6.4, "Resizing File System Cache"

4.6.2 Enabling File System Cache

The cache.conf parameter PlsqlCacheEnable enables mod_plsql caching. For maximum performance, enable PlsqlCacheEnable by setting the value of this parameter to On.

4.6.3 Configuring File System Cache to Reside on a Faster File System

This section describes how to configure a File System Cache to reside on a separate disk. When you use File System Cache and store the cache on a faster separate disk, performance should improve for all types of Web applications using File System Cache, including OracleAS Portal and generic PL/SQL based Web applications.

When you configure File System Cache, the cache can reside either on a separate physical disk or in a memory disk.

To set up a File System Cache on a separate disk:

1. Assume that the file system for the cache resides at the location:

   On UNIX: /u01/cache

   On Windows: E:\cache

2. Update the file:

   On UNIX: ORACLE_HOME/Apache/modplsql/conf/cache.conf

   On Windows: ORACLE_HOME\Apache\modplsql\conf\cache.conf

3. Change the cache parameter PlsqlCacheDirectory:

   On UNIX: PlsqlCacheDirectory /u01/cache

   On Windows: PlsqlCacheDirectory E:\cache

4.6.4 Resizing File System Cache

This section covers the following topics:

• Setting the Total Cache Size with PlsqlCacheTotalSize
• Setting the Days of Aging for Cache with PlsqlCacheMaxAge
• Setting the Maximum File Size for a Cache File with PlsqlCacheMaxSize

4.6.4.1 Setting the Total Cache Size with PlsqlCacheTotalSize

The default installation sets the mod_plsql file system cache size to 2097152 bytes (20 Megabytes). If your PL/SQL application does not make use of the OWA_CACHE packages, or uses them to cache small amounts of content, then the default setting should be sufficient. If your PL/SQL application caches a lot of content in the mod_plsql file system cache, you should consider specifying a higher value.

To control the cache size, set the PlsqlCacheTotalSize parameter in the file cache.conf. On UNIX systems, this file is located under ORACLE_HOME/Apache/modplsql/conf directory. On Windows systems, this file is located under ORACLE_HOME\Apache\modplsql\conf.

You need to set the cache size high enough to achieve a high cache hit ratio. Try to set the cache size large enough so that frequently accessed content stays cached. It is also important to limit the amount of disk space, so that the cache size does not grow too large. Correct tuning for the cache size provides enough cache to hold all frequently accessed content while preventing the cache size from growing too large, since a very large cache is inefficient to search.

The value for PlsqlCacheTotalSize is specified as a number of bytes. 1MB equals 1048576 bytes. This setting is a soft limit on the amount of cache allocated. In some cases, the cache size may grow beyond this limit until the next cleanup operation. Therefore, the hard limit on the cache size is the underlying physical hard disk size. When this limit is reached, no cache content can be written out to disk until space is available.

To determine a reasonable cache size, do the following:

1. Turn on mod_plsql performance logging by setting the LogLevel in httpd.conf to the info level to enable mod_plsql logging.
2. Monitor the error_log on a daily basis. On UNIX systems, the default directory for the error log is: ORACLE_HOME/Apache/Apache/log. On Windows systems, the default directory is: ORACLE_HOME\Apache\Apache\log.

The mod_plsql error_log entries have the form:

[info] mod_plsql: cachecleanup deleted=2571 max_age=96,2178852b  
kept=1042,25585368b time=128s limit=25600000b

where:

deleted is the number of cache files that got deleted during the cleanup process.

max_age is the number of cache files and total size that got deleted because they haven't been used for some time.

kept is the number of cache files and total size that was kept after the cleanup process.

time is the amount of time to perform the cleanup.

limit is the total cache size. This is the value of the PlsqlCacheTotalSize setting.

Interpret the entries in the error log as follows:

• If a high number of files are being deleted when compared to the number of files that were kept, this is a clear indication that your cache size is too small. You probably need to increase the size of the cache.
• If a low number of files being deleted when compared to the number of files that were kept is observed, this is an indication that your cache size is probably too big. If you have enough disk space, you can chose to leave it as it or you can decrease the size of the cache to reclaim some disk space.

4.6.4.2 Setting the Days of Aging for Cache with PlsqlCacheMaxAge

Using the PlsqlCacheMaxAge parameter, you can control the "staleness" of cache content. The value for parameter is specified in units of days. The default value for this parameter is 30 (days). This means cache content is kept in the cache if it is less than 30 days old. After 30 days, the content is considered for deletion during the cleanup process.

The max_age information in mod_plsql error_log shows cache file aging information. If your site is a highly dynamic site, it would make sense to configure this setting to a lower value, since the older cache content will usually not be used again and, therefore, the lower value does not affect the cache hit ratio. If the site contains many static pages, it would make sense to increase the value of PlsqlCacheMaxAge so that the cleanup process does not deliberately delete the cache content.

4.6.4.3 Setting the Maximum File Size for a Cache File with PlsqlCacheMaxSize

Using the PlsqlCacheMaxSize parameter, you can specify the maximum size for individual files in the cache. Using this parameter prevents the case in which one cache file fills up the entire cache.

The default value for this parameter is 1048576 (bytes). In general, set this parameter to a value that represents about 1-3% of the total cache size.

4.6.5 Configuring Cache Cleanup

The cache cleanup parameter determines the frequency in which the File System Cache is examined and, if necessary, cleaned up. The cache cleanup parameter, PlsqlCacheCleanupTime is specified in the cache.conf file. The frequency can be set to daily, weekly, or monthly. When specifying weekly cleanup, it is possible to specify the day of the week and the time of the day.

The default mod_plsql setting of PlsqlCacheCleanupTime is daily at 11PM local time. Therefore, by default, every night at 11PM, the cleanup routine runs. When you select the monthly frequency, the cleanup occurs on the first Saturday of each month.

Configuring this parameter correctly is important since cleaning up too often can result in a lower cache hit ratio and when cleaning does not occur often enough, the cache's disk usage may be excessive.

Monitor the cleanup activities using the entries in the mod_plsql error_log; then tune the cleanup parameter, PlsqlCacheCleanupTime by analyzing the entries.

[info] mod_plsql: cachecleanup deleted=2571 max_age=96,2178852b  
kept=1042,25585368b time=128s limit=25600000b

Note the following:

• Seeing a large number for the cleanup time can be an indication that the cleanup frequency is set too low. When the log indicates that the cleanup operation is busy examining or deleting many cache files, increasing the cleanup frequency should decrease the time spent in the cleanup operation.
• If a high number files are being deleted during the cleanup operation because of "staleness", this is an indication that the cleanup frequency is too low. In this case, increase the frequency so that the cleanup can actively delete "stale" cache content more frequently.

4.7 Oracle HTTP Server Directives

To improve PL/SQL performance in Oracle HTTP Server, you need to tune the Oracle HTTP Server directives appropriately for your configuration.