| | h2. AIX V5.3 initial tuning recommendations |
| | |
| | (!) The contents of this web page solely reflect the personal views of the authors and do not necessarily represent the views, positions, strategies or opinions of IBM or IBM management. Please use the !http://www.ibm.com/developerworks/wikis/images/icons/add_comment_16.gif! +Add Comment+ link at the bottom of the page to provide feedback. *Note:* Until you log in (using the link in the upper right corner of this web page), you will not see the !http://www.ibm.com/developerworks/wikis/images/icons/add_comment_16.gif! +Add Comment+ link and you can not add a comment. If you do not already have an IBM ID, use the +Register Now+ link on the sign in page to obtain one. Registration is quick and easy. |
| | {note:title=Note} |
| | There is no substitute for hands-on training with an instructor to learn the basics before attempting the first AIX implementation on a production system. See the [IBM Training|http://www-304.ibm.com/jct03001c/services/learning/ites.wss/tp/en?pageType=page&c=a0011023] web page. In the US, see the [IBM Training AIX 5L System Administrators|http://www-304.ibm.com/jct03001c/services/learning/ites.wss/us/en?pageType=page&c=a0000214] web page to select a course. |
| | {note} |
| | |
| | This web page is meant to supplement, *not replace*, other documentation which has been published: |
| | \\ |
| | * The [??Performance tuning??|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.prftungd/doc/prftungd/prftuning.htm] article in the [??AIX V5.3 Performance management??|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.prftungd/doc/prftungd/performance_management-kickoff.htm] manual |
| | * AIX tuning recommendations for Oracle: |
| | *# [??Oracle Database 11g and Oracle RAC 11g on IBM AIX: Tips and Considerations??|http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101176] |
| |  | *# [??Oracle Database 10gR2 and Oracle RAC 10gR2 on IBM AIX: Tips and Considerations??|http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101089] |
| | *# [??Oracle 9i & 10g on IBM AIX 5L: Tips & Considerations??|http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP100556] |
 | | *# [??Oracle DB & RAC 10gR2 on IBM AIX 5L: Tips and Considerations??|http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101089] |
| | *# [??Implementing Oracle 10g RAC with ASM on AIX??|http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP100708] white paper |
| | *# [??Oracle 10g RAC Deployment on AIX Using VERITAS Storage Foundation for Oracle RAC??|http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP100930] white paper |
| | *# [??Oracle Architecture and Tuning on AIX??|http://www.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP100883] white paper |
| | *# [??Tuning IBM AIX 5L for an Oracle Database??|https://www-304.ibm.com/jct09002c/partnerworld/wps/servlet/ContentHandler/VPAA-6V2C4E] white paper |
| | | *# [??Tuning IBM AIX 5.3 and AIX 6.1 for Oracle Database??|http://www.ibm.com/partnerworld/wps/servlet/ContentHandler/whitepaper/aix/oracle/performance_analysis] white paper |
| | |
| | h3. AIX V5.3 initial tuning |
| | |
| | The vast majority of AIX kernel tuning parameters are set appropriately by the AIX installation process. There are, however, a few tuning parameters which should be changed from the default values as soon as AIX is installed. |
| | |
| | Tuning changes should be generally be made only with a specific objective in mind and in reaction to a performance issue which it seems likely that AIX kernel tuning can address. *The changes recommended here are an exception to that general rule.* |
| | {info:title=vmo tunables} |
| | The AIX lru_file_repage kernel tuning parameter was introduced in AIX V5.2 ML5 and AIX V5.3 ML2. The default setting for lru_file_repage is 1 \(on\). Before lru_file_repage was introduced, it was often necessary to lower maxperm/maxclient to get good performance, particularly when running an Oracle RDBMS workload. Now that lru_file_repage is available, when installing AIX V5.3 at recent Technology Levels it is best to set: |
| | {noformat} |
| | lru_file_repage = 0 |
| | maxclient% = 90 |
| | maxperm% = 90 |
| | minperm% = 5 |
| | {noformat} |
| | The settings of these parameters can be changed with the [{{vmo}}|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.cmds/doc/aixcmds6/vmo.htm] command. When making these (and most) tuning changes, the best practice is to use the -p flag on the [{{vmo}}|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.cmds/doc/aixcmds6/vmo.htm] command, which causes tuning parameters to be changed on the running AIX image and configures AIX so that parameters will be set as specified each time AIX boots. |
| | |
| | See the [Optimizing AIX performance tuning|http://www.ibm.com/developerworks/aix/library/au-aix6tuning/index.html#table1] web page on IBM developerWorks for more information. |
| | |
| | Subsequent AIX performance monitoring and analysis may suggest changing other [{{vmo}}|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.cmds/doc/aixcmds6/vmo.htm] parameters, but it is best to start with other vmo tunables at their default values unless other AIX V5.3 initial tuning recommendations come from some authoritative source. |
| | {note:title=Please note} |
| | Default values change from one Technology Level to another, but when AIX is upgraded to a new Technology Level, kernel tuning parameters are not changed by the upgrade, so kernel tuning parameters might not be set to the appropriate default values. Please examine {{/etc/tunables/nextboot}} to determine which kernel tuning parameters will be changed from the default value the next time AIX is rebooted. Set parameters to default values unless another setting is recommended or has been confirmed empirically to work better than the default. |
| | {note} |
| | {info} |
| | {info:title=TCP tuning} |
| | The tcp_sendspace, tcp_recvspace, and sb_max tunables should be set as specified in the table in the [??TCP streaming workload tuning??|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.prftungd/doc/prftungd/tcp_streaming_workload_tuning.htm] article in the [??AIX V5.3 Performance management??|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.prftungd/doc/prftungd/performance_management-kickoff.htm] manual, unless there is good reason to believe that the TCP workload will be a [request and response workload|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.prftungd/doc/prftungd/tcp_request_response_workload_tun.htm] rather than streaming. For example, as of 2/25/2008 the recommended values in the table for a Gigabit Ethernet adapter with a 1500-byte MTU size (not jumbo frames) are tcp_sendspace=131072, tcp_recvspace=65536, and sb_max=1048576 (recommended in a footnote to the table). |
| | |
| | In this case (and many others), the table recommends setting rfc1323=0, but rfc1323=1 is now recommended in all cases. |
| | |
| | (/) Use the {{[entstat|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.cmds/doc/aixcmds2/entstat.htm]-d ent0 \| grep -i media}} commands to display the media speed at which a given physical adapter (ent0 in this example) is running. For a virtual adapter, use the media speed of the physical adapter to which the virtual adapter is mapped. |
| | {info} |
| | {info:title=UDP tuning} |
| | The udp_sendspace and udp_recvspace tunables should be set as recommended in the [??UDP tuning??|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.prftungd/doc/prftungd/udp_tun.htm] article in the [??AIX V5.3 Performance management??|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.prftungd/doc/prftungd/performance_management-kickoff.htm] manual. As of 2/25/2008, suggested starting values in the article are udp_sendspace=655360 and udp_recvspace=65536. |
| | {info} |
| | {info:title=I/O pacing} |
| | The [Disk I/O pacing|http://publib.boulder.ibm.com/infocenter/systems/scope/aix/topic/com.ibm.aix.prftungd/doc/prftungd/disk_io_pacing.htm] article states that the default value for the high water mark (maxpout) is 8193 and for the low water mark (minpout) is 4096. The defaults are recommended. {color:red}Please note that the following HACMP smit panel contextual help is misleading:{color} |
| | {noformat} |
| | +--------------------------------------------------------------------------+ |
| | | CONTEXTUAL HELP | |
| | | | |
| | | Press Enter or Cancel to return to the application. | |
| | | | |
| | | For HACMP ES, recommended values are a high water mark of 33 | |
| | | and a low water mark of 24. | |
| | | | |
| | | General information: | |
| | | Indicates the high water mark for pending write-behind I/Os per file. | |
| | | This attribute, along with the low water mark for pending | |
| | | write-behind I/Os per file attribute, may be used in a | |
| | | multiprogramming environment to balance and control the I/O activity | |
| | | associated with file system write behind. The high water mark | |
| | | specifies the number of write-behind I/Os per file at which | |
| | | additional blocks write. The low water mark specifies the number of | |
| | | pending write-behind I/Os per file at which previously blocked writes | |
| | | are allowed to proceed. The possible values for the high water mark | |
| | | range from 0 to 32767. The possible values for the low water mark | |
| | | range from 0 to 32766. If set to a value other than zero, the low | |
| | | water mark value must be less than the high water mark value. If the | |
| | | high water mark is set at 0, the low water mark must also be 0. When | |
| | | both values are 0, no controls are placed on file system write-behind | |
| | | activity. By default, both values are 0. A high water mark value of | |
| | | 32 and a low water mark of 24 may be appropriate in environments with | |
| | | high multiprogramming levels. | |
| | | | |
| | | F1=Help F2=Refresh F3=Cancel | |
| | | F8=Image Enter=Do | |
| | +--------------------------------------------------------------------------+ |
| | {noformat} |
| | {color:red}As of 7/2009, there is a plan to update the contextual help cited above to provide better advice.{color} |
| | {info} |
| | |
| | (i) Once initial tuning changes have been implemented, remember that tuning of any operating system (including AIX) is an interative process. That is, performance requirements should be identified, current performance should be assessed and, if appropriate, a small set of changes should be made. The effects of those changes should be assessed to determine if the changes caused any improvement. Changes which don't result in an improvement should generally be backed out. Changes which do result in an improvement should be preserved as the basis for further tuning. For more information, see [??The performance-tuning process??|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.prftungd/doc/prftungd/intro_perf_tuning_process.htm] article in the [??AIX V5.3 Performance management??|http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.aix.prftungd/doc/prftungd/performance_management-kickoff.htm] manual. Please note that it is often necessary to change AIX tuning in tandem with other tuning (eg, Oracle RDBMS memory use tuning). |
