Optimizing AIX 6.1 performance tuning
Tuning methodology, general enhancements, monitoring, and tuning
AIX® 6.1 has many important innovations and improvements, including:
- Virtualization—Features such as workload partitioning and Live Application Mobility
- Security—Features such as encrypted file systems and trusted AIX and role-based access control (RBAC)
- Availability—Features such as AIX concurrent updates and dynamic tracing
- Manageability—Features such as the new Systems Director Console for AIX and Workload Partition Manager
It also has support for POWER6 performance innovations, such as advanced simultaneous multithreading (SMT), shared dedicated processors, and variable page size. It's important to fully understand which innovations and enhancements are more the reflection of the POWER6, AIX 6.1, or a combination of both. For example, purely from an operating system perspective, AIX improves on the older defaults tunables for:
While there are some real performance enhancements in AIX 6.1, such as improvements in I/O pacing and AIX's implementation of AIO servers, there is nothing breathtakingly different. In fact, there were more performance changes made from AIX 5.1 to AIX 5.2 and 5.2 to 5.3 (including new monitoring tools, tuning tools, new tunables that changed how you set Virtual Memory Manager (VMM) settings, and concurrent I/O improvements) than you will see moving from AIX 5.3 to AIX 6.1. In AIX 6.1, all the tuning commands remain the same, except for the ones that have been taken away (I.E. and aioo), and there are no new monitoring tools either. Other changes reflect updates that were made to the utilities that reflect support for other workload partitioning innovations, including:
Workload partitions (WPARs) allow for the usage of separate virtual partitions within one AIX image. It is more of a compliment to logical partitions (LPARs) rather than a replacement to them. WPARs actually run inside of LPARs and are very similar in concept to Solaris containers. I've built WPARs in less than 15 minutes; in fact, you'll do some of your analysis inside WPARs so that you can actually view some of the updated tools that now support WPARs. Note that WPARs are only required in AIX 6.1 and a POWER6 is not necessary. Some commands also run differently or not run at all within WPARs—a few of them will be discussed where applicable. Other AIX 6.1 performance improvements include unique tunable documentation and restricted tunables.
This section summarizes a tuning methodology that you should use to tune your systems. You don't necessarily have to use this one; there are many others, but you should be working with some kind of tuning methodology or structured program.
The five-step methodology consists of:
- Establishing a baseline
- Stress testing and monitoring
- Identification of bottleneck
- Tuning bottleneck
- Repeat (starting with the second step)
Let's look at the steps in more depth:
- Establishing a baseline—Before you tune or even start monitoring, you must establish a baseline. The baseline is a snapshot of what the system looks when it is performing well. This baseline should not only capture performance-type statistics, but it should also document the actual configuration of your system (amount of memory, CPU, and disk). This is important, because you need to know what a system that performs well looks like prior to troubleshooting performance that users might have complained about. Doing so helps you translate the data into a service level agreement (SLA) in which the customer signs off on.
- Stress testing and monitoring—Here is where you monitor and stress your systems at peak workloads. You should use several monitoring tools here to help validate your findings. The monitoring section is critical, as you cannot effectively tune anything without having an accurate historical record of what has been going on in your system, particularly during periods of stress.
- Identification of bottleneck—The objective of stressing and monitoring your system is to determine the bottleneck. You cannot provide the correct medicine without the proper diagnosis. If the system is in fact CPU bound, then you can run additional tools, such as trace, curt, splat, tprof, and ps to further identify the actual processes that are causing the bottleneck.
- Tuning bottleneck—After you've finally identified the
bottleneck, it is time to tune it. The tuning that you do is dependent on the
bottleneck, for example, CPU, virtual memory, or I/O. Some subsystems lend
themselves more to active tuning, such as virtual memory and using
vmo, while the "cure" for CPU-type bottlenecks is usually managing your workload more efficiently or assigning (using dynamic logical partitioning (DLPAR), uncapped partitions, or Partition Load Manager) more resources to your system.
- Repeat—Time to go through this process again, starting from the second step of the process. Only by repeating your tests and consistently monitoring your systems can you determine if your tuning has really made an impact.
This section discusses the changes made in the virtual memory subsystem.
Through the years, many have complained about the default parameters of AIX. The
complaints have not just reflected the reality of what most users run on top of
AIX, for example, mission-critical database applications like Oracle. Because of
this, systems administrators have had to change settings on many
subsystems—most notably with virtual memory (in other words,
IBM® has listened and changed these parameters to reflect that reality. It
should also be noted that you should not rely on these settings exclusively, and
you should always check with your independent software vendor (ISV) to validate
what their recommended settings are for AIX 6.1 and make the changes, accordingly.
Let's look at a side-by-side view of
for AIX 5.3 and AIX 6.1 in Table 1. You will see that some
parameters are now in restricted mode. The most important changes were made to
address paging issues, where database servers frequently page out computational
pages, even though the system has enough free memory. In prior articles on tuning
(see Related topics), it was recommended to change those
parameters to defaults fairly close to what was indicated on the table. The
changes are indicated in the AIX 5.3 tuning recommendation column. Many tunables
are now also classified differently as restricted in an attempt to discourage
changing certain parameters. The net is that only 29
vmo tunables can now be changed without a firm warning;
30 are now deemed restricted tunables, which IBM officially states should not be
modified unless instructed by "IBM support professionals." IBM has made this
change in an attempt to discourage junior administrators from changing certain
parameters that were deemed critical enough to be classified as restricted.
Table 1. List of
vmo parameters and tuning recommendations
|Type||AIX 5.3||AIX 5.3 tuning change recommendations||AIX 6.1||Restricted|
|cpu_scale_memp = 8||8||8||Yes|
|data_stagger_interval = 161||161||161||Yes|
|defps = 1||1|
|force_relalias_lite = 0||0||0|
|framesets = 2||2||2||no longer available||Yes|
|htabscale = n/a||n/a||n/a||Yes|
|kernel_heap_psize = 4096||4096||4096|
|kernel_psize = 4096||4096||65536||Yes|
|large_page_heap_size = 0||0||0||Yes|
|lgpg_regions = 0||0||0|
|lgpg_size = 0||0||0|
|low_ps_handling = 1||1||0|
|lru_file_repage = 1||1||0||0||Yes|
|lru_poll_interval = 10||10||10||Yes|
|lrubucket = 131072||131072||131072||Yes|
|maxclient% = 80||80||90||90||Yes|
|maxfree = 1088||1088||1088|
|maxperm = 598781||598781||442683|
|maxperm% = 80||80||90||90||Yes|
|maxpin = 635249||635249||423471|
|maxpin% = 80||80||80|
|mbuf_heap_psize = 65536||65536||64436||Yes|
|memory_affinity = 1||1||1||Yes|
|memory_frames = 786432||786432||524288|
|memplace_data = 2||2||2|
|memplace_mapped_file = 2||2||2|
|memplace_shm_anonymous = 2||2||2|
|memplace_shm_named = 2||2||2|
|memplace_stack = 2||2||2|
|memplace_text = 2||2||2|
|memplace_unmapped_file = 2||2||2|
|mempools = 5||5|
|minfree = 960||960||960|
|minperm = 149694||149694||14756|
|minperm% = 20||20||5||3|
|nokilluid = 0||0||0|
|npskill = 3072||3072||3072|
|npsrpgmax = 24576||24576||8192||Yes|
|npsrpgmin = 18432||18432||6144||Yes|
|npsscrubmax = 24576||24576||8192||Yes|
|npsscrubmin = 18432||18432||6144||Yes|
|npswarn = 12288||12288||12288|
|num_spec_dataseg = 0||0||0||Yes|
|numpsblks = 393216||393216||393216|
|page_steal_method = 0||0||1||Yes|
|pagecoloring = n/a||n/a|
|pinnable_frames = 675516||675516||349814|
|pta_balance_threshold = n/a||n/a|
|relalias_percentage = 0||0||0|
|rpgclean = 0||0||0||Yes|
|rpgcontrol = 2||2||2||Yes|
|scrub = 0||0||0|
|scrubclean = 0||0||0||Yes|
|soft_min_lgpgs_vmpool = 0||0|
|spec_dataseg_int = 512||512||512||Yes|
|strict_maxclient = 1||1||1||Yes|
|strict_maxperm = 0||0||0||Yes|
|v_pinshm = 0||0||0|
|vm_modlist_threshold = -1||-1||-1||Yes|
|vmm_fork_policy = 1||1||1||Yes|
|vmm_mpsize_support = 1||1||2||Yes|
provides all parameters, including restricted ones. Most of the content here is
stripped out—only a small sampling is provided that includes the
restricted section (see Listing 1).
Listing 1. Using the
vmo to view all the tunable values
lpar9ml162f_pub[/] > vmo -F -a force_relalias_lite = 0 vmm_default_pspa = -1 ##Restricted tunables maxperm% = 90
Even restricted tunables can be changed. When you do make a change, you just receive a stern warning (see Listing 2).
Listing 2. Adjusting tunables warning
lpar9ml162f_pub[/] > vmo -o maxperm%=99 Setting maxperm% to 99 Warning: a restricted tunable has been modified
When changing the parameter after a reboot, you'll get a further rebuke and be asked to actually confirm whether or not you want do to this. You'll have to physically type in "yes" (see Listing 3).
Listing 3. Restricted tunables—answering yes to change
lpar9ml162f_pub[/] > vmo -p -o maxperm%=99 Modification to restricted tunable maxperm%, confirmation required yes/no yes Setting maxperm% to 99 in nextboot file Setting maxperm% to 99 Warning: a restricted tunable has been modified lpar9ml162f_pub[/] >
The most important out-of-the-box performance values include new values for:
This is based on changes that first appeared during AIX 5.3, when you no longer
had to turn off
maxclient. The new recommendation (now incorporated as
the default value in AIX 6.1) is to turn off the repage ratio check,
lru_file_repage, which ensures that working storage is
not paged and to consider only file paging. In AIX 6.1, the VMM replacement
default is changed to utilize up to 90 percent of its real memory for file
caching, favoring computational pages over file pages. Unless the amount of active
virtual memory exceeds 97 percent of the size of real memory,
minperm is reduced to three percent (see
Listing 4) to ensure that computational pages will not be
to three percent
# vmo -o minperm%=97 Value of the tunable minperm% cannot be changed in a WPAR
Some changes will not work in WPARs. WPARs are a subset of an LPAR, but they are still part of the single operating system image.
Another important change includes VMM dynamic variable page size support (VPSS).
Pages are defined as fixed length data blocks and held in virtual memory. In AIX
6.1 (on POWER6 processors only), VMM can now dynamically use the larger page size
based on application memory usage, which should substantially improve performance.
This feature is completely transparent to applications. AIX only uses the larger
pages sized if it doesn't result in increasing process memory usage. The use of
larger pages increases performance because fewer hardware address translations
would need to be made. It only supports doing this on working storage memory, not
persistent storage. The new parameters are
vmm_default_pspa (it also works in conjunction with the
lsattr output illustrating processor architecture
lpar9ml162f_pub[/usr/samples/tcpip] > lsattr -El proc0 Frequency 1654344000 Processor Speed False smt_enabled true Processor SMT enabled False smt_threads 2 Processor SMT threads False state enable Processor state False type PowerPC_POWER5 Processor type False lpar9ml162f_pub[/usr/samples/tcpip] >
Let's view the setting for tunable settings for VPSS.
Listing 6. VPSS automatically disabled on a POWER5 system
lpar9ml162f_pub[/usr/samples/tcpip] > vmo -a | grep pspa vmm_default_pspa = -1
This section highlights changes made in the CPU subsystem.
schedo can also be used to tune certain memory
facilities, such as paging load control, it is mostly there for CPU scheduler
tuning. In AIX 6.1, only 27 of the 42 tunables are restricted, leaving 15 tunables
that can be modified without explicit warnings. Though some defaults were changed,
there were no substantial changes that were made with respect to CPU monitoring
and tuning on AIX 6.1 (see Listing 7).
Listing 7. Restricted tunables
lpar9ml162f_pub[/] > schedo -F -a ##Restricted tunables %usDelta = 100 allowMCMmigrate = 0 fast_locks = n/a hotlocks_enable = 0 idle_migration_barrier = 4 krlock_confer2self = 1 krlock_conferb4alloc = 1 krlock_enable = 1 krlock_spinb4alloc = 1 krlock_spinb4confer = 1024 n_idle_loop_vlopri = 100 search_globalrq_mload = 256 search_smtrunq_mload = 256 setnewrq_sidle_mload = 384 shed_primrunq_mload = 64 sidle_S1runq_mload = 64 sidle_S2runq_mload = 134 sidle_S3runq_mload = 134 sidle_S4runq_mload = 4294967040 slock_spinb4confer = 1024 smt_snooze_delay = 0 smtrunq_load_diff = 2 v_exempt_secs = 2 v_min_process = 2 v_repage_hi = 0 v_repage_proc = 4 v_sec_wait = 1 lpar9ml162f_pub[/] >
AIX 6.1—Disk I/O
This section gives an overview of the changes that were made in the disk I/O subsystem.
In AIX 6.1, of the 48 tunables you can control with
ioo, 27 of them are now restricted, leaving 21 tunables
that can be modified without explicit warnings. The most important changes impact
I/O pacing and AIO dynamic tunables (see Listing
Listing 8. Restricted
# ioo -F -a ##Restricted tunables aio_fastpath = 1 aio_fsfastpath = 1 aio_kprocprio = 39 aio_multitidsusp = 1 aio_sample_rate = 5 aio_samples_per_cycle = 6 j2_maxUsableMaxTransfer = 512 j2_nBufferPerPagerDevice = 512 j2_nonFatalCrashesSystem= 0 j2_syncModifiedMapped = 1 j2_syncdLogSyncInterval = 1 jfs_clread_enabled = 0 jfs_use_read_lock = 1 maxpgahead = 8 maxrandwrt = 0 memory_frames = 524288 minpgahead = 2 numclust = 1 numfsbufs = 196 pgahd_scale_thresh = 0 posix_aio_fastpath = 1 posix_aio_fsfastpath = 1 posix_aio_kprocprio = 39 posix_aio_sample_rate = 5 posix_aio_samples_per_cycle = 6 pv_min_pbuf = 512 sync_release_ilock = 0
Changes were made to the Enhanced Journaling File System (JFS2) that now allow
you to mount a JFS2 f/s without logging. While this can substantially improve
performance, it is not recommended you do this. If you need to recover your data,
you have to use the dreaded
fsck, which has been pretty
much banished from memory since journaling file systems. Some useful circumstances
can include restoring data from backups and saving time during an activity where
you might have a very small window and availability is not a concern.
The target software driver can now be used over a gigabyte Ethernet adapter, which should clearly improve performance in this type of environment. The target driver exports local disks or logical volumes to iSCSI initiators that connect to AIX using the iSCSI protocol. The proliferation of iSCSI is a viable alternative over Fiber-based storage, making this an important enhancement.
Disk I/O pacing is a mechanism that allows you to limit the number of pending I/O
requests to a file. What this does is prevents disk I/O-intensive processes
(usually in the form of large sequential writes) from exhausting the CPU. AIX 6.1
enables I/O pacing by default. In AIX 5.3, you needed to explicitly enable this
feature. The way it does this is by setting the
maxput parameters to 4096 and 8193, respectively:
lpar9ml162f_pub[/] > lpar9ml162f_pub[/] > lsattr -El sys0
output illustrating new I/O pacing settings
Asynchronous I/O (AIO)
AIO is an AIX software subsystem that allows processes to issue I/O operations
without waiting for I/O to finish. Because both I/O operations and applications
processing run concurrently, they essentially run in the background and improve
performance. This is particularly important in a database environment. There are
two types of AIX subsystems: Legacy AIO and POSIX AIO. The differences between
them involve different parameter passing at the application layer. In other words,
the developers pick the implementation that the application uses. Regardless of
which subsystem is chosen, both run concurrently on AIX. In AIX 5L™, if
applications used AIO, the subsystem would need to be explicitly activated in the
autoconfig parameter. The system also requires a reboot
because the kernel extensions need to be loaded. In fact, any release prior to AIX
5.3 TL5 required reboots if there were any changes to the either of the following tunables:
In AIX 5.3, you also had the
aioo command, which
allowed you to make these changes dynamically without a reboot (decreasing
required reboots). It should be noted that the command did not change the Object
Data Manager (ODM) attributes, meaning they would not be persistent across a
In AIX 6.1, the tunables
fsfastpath are now restricted tunables and set to 1 by
default. It has the following effect on the tunables (see
fastpath: AIO requests that raw logical volumes get passed directly to the disk layer.
fsfastpath: AIO requests that files opened with concurrent I/O on JFS2 get passed directly to LVM or disk.
Listing 9. Restricted tunables
##Restricted tunables aio_fastpath = 1 aio_fsfastpath = 1
Further, AIO subsystems are now loaded by default and not activated. They are
automatically started at the time when the application initiates the AIO I/O
requests. There is no more
aioo command (what a short
life span) and these tunables are now used only with
ioo (see Listing 10).
Listing 10. AIX old method (AIX 5.3)
lpar9ml24f_pub[/] > # aioo -a minservers = 1 maxservers = 1 maxreqs= 4096 fsfastpath = 0
Listing 11 provides the new method for AIX 6.1.
Listing 11. AIO new method (AIX 6.1)
lpar9ml162f_pub[/] > ioo -a | grep active aio_active = 0 posix_aio_active =
It's also worth noting that there are no more AIO devices in the ODM.
As you can see, two new parameters have also been added to
posix_aix_active. These can only be changed by AIX, and
they are set to 1 only when AIO kernel extensors are used and pinned. If you like
grep, you won't find any more AIO servers. You'll now
these are the kernel processes that manage the AIO subsystems for Legacy and
POSIX. As a result of this change, there is less pinned memory and fewer processes
that are running on the system; both have positive affects on overall systems
performance (see Listing 12).
Listing 12. New AIO kernel processes
lpar9ml162f_pub[/] > pstat -a | grep aio 39 a 2704e 1 2704e 0 0 1 aioLpool 40 a 28050 1 28050 0 0 1 aioPpool lpar9ml162f_pub[/] >
parameters, as they relate to AIO servers, are now tuned per each CPU tunable.
Changing these values will not result in changes to the number of available
servers on the system; the amount is dependent on the number of concurrent I/O
requests (see Listing 13).
Listing 13. Changes in default parameters for
lpar9ml162f_pub[/] > ioo -a | grep minservers aio_minservers = 3 posix_aio_minservers = 3 lpar9ml162f_pub[/] > ioo -a | grep maxservers aio_maxservers = 30 posix_aio_maxservers = 30 lpar9ml162f_pub[/] >
It's also worth noting that there are no more AIO devices in the ODM.
AIX 6.1—Network and Network File System (NFS) I/O
This section outlines changes that were made in the Network I/O subsystem, including NFS.
Of the 133
no tunables, only five have been classified
as restricted (see Listing
Listing 14. Restricted tunables
#no -F -a ##Restricted tunables extendednetstats = 0 inet_stack_size = 16 net_malloc_police = 16384 pseintrstack = 24576 use_isno = 1 lpar9ml162f_pub[/] >
A new network caching daemon has also been introduced to improve performance when resolving using DNS.
This daemon can be started from the System Resource Controller (SRC).
It's main configuration file is in /etc/netcd.conf, and you can copy the one in
/usr/samples/tcpip to /etc and use that as a template. The command used to manage
the daemon is
netcdctrl. With this command, you can
dump the cache contents to a file, display statistics of cache usage, flush the
cache table, and change the logging level of the daemon.
Listing 15 displays the stats.
Listing 15. Displaying caching stats with
lpar9ml162f_pub[/usr/samples/tcpip] > netcdctrl -t dns -e hosts -s /tmp/netcd.stats lpar9ml162f_pub[/usr/samples/tcpip] > more /tmp/netcd.stats CACHE dns, hosts, name Hash index : 0, Max number of entries : 0, Current number of entries : 0 Hash index : 1, Max number of entries : 0, Current number of entries : 0 Hash index : 2, Max number of entries : 0, Current number of entries : 0 Hash index : 3, Max number of entries : 0, Current number of entries : 0
Regarding the /etc/netsvc.conf file, nothing has changed; this file is still necessary in determining the order of resolving.
Of the 24 NFS tunables, 21 have been classified as restricted (see Listing 16).
Listing 16. Restricted tunables
#nfso -a ##Restricted tunables lockd_debug_level = 0 nfs_allow_all_signals = 0 nfs_auto_rbr_trigger = 0 nfs_dynamic_retrans = 1 nfs_gather_threshold = 4096 nfs_iopace_pages = 0 nfs_max_threads = 3891 nfs_repeat_messages = 0 nfs_socketsize = 600000 nfs_tcp_duplicate_cache_size = 5000 nfs_tcp_socketsize = 600000 nfs_udp_duplicate_cache_size = 5000 nfs_v2_pdts = 1 nfs_v3_pdts = 1 nfs_v4_pdts = 1 nfs_v2_vm_bufs = 10000 nfs_v3_vm_bufs = 10000 nfs_v4_vm_bufs = 10000 statd_debug_level = 0 statd_max_threads = 50 udpchecksum = 1 lpar9ml162f_pub[/] >
The only noteworthy change here is that RFC1323 (on the TC/IP stack) is now
enabled by default, which allows TCP connections to use the TCP scaling window for
any NFS connections. Further, the default number of
biod daemons has now increased to 32 for each NFS V3
AIX 6.1 provides many innovations, such as the use of WPARs and live application availability. It also provides for many changes with respect to performance tuning. Some of the more important changes include improving default parameters to more accurately reflect real-world data processing, unique tunable documentation, the incorporation of restricted tunables, and enhancements to I/O pacing and AIO servers. AIX 6.1 is simply wonderful, but also remember that in the world of IBM, AIX 6.1 is the new 6.0. If you recall, AIX 5.2 was a lot better and more stable than AIX 5.1. Consider everything before doing an operating system upgrade, but certainly don't be afraid of moving over if your application fully supports it and you feel the innovations are important enough in your environment to make the move.
- Use RSS feed to request notification for the
upcoming articles in the following series:
- Optimizing AIX 5L™ performance: Tuning disk performance (Ken Milberg, developerWorks, November 2007)
- Optimizing AIX 5L™ performance: Tuning disk performance (Ken Milberg, developerWorks, July 2007)
- Optimizing AIX 5L performance: Tuning your memory settings (Ken Milberg, developerWorks, June 2007)
- Optimizing AIX 5L performance: Monitoring your CPU (Ken Milberg, developerWorks, April 2007)
- AIX memory affinity support: Learn more about affinity support from the IBM System p™ and AIX InfoCenter.
- IBM Redbooks: Database Performance Tuning on AIX is designed to help system designers, system administrators, and database adminsitrators design, size, implement, maintain, monitor, and tune a Relational Database Management System (RDMBS) for optimal performance on AIX.
- "Processor Affinity on AIX" (developerWorks, November 2006): Using process affinity settings to bind or unbind threads can help you find the root cause of troublesome hang or deadlock problems. Read this article to learn how to use processor affinity to restrict a process and run it only on a specified central processing unit (CPU).
- IBM Redbooks: For a comprehensive guide about the performance monitoring and tuning tools that are provided with AIX 5L Version 5.3, read AIX 5L Practical Performance Tools and Tuning Guide.
- IBM Redbooks: AIX 5.3 Performance Management Guide provides application programmers, customer engineers, system engineers, system administrators, experienced end users, and system programmers with complete information about how to perform such tasks as assessing and tuning the performance of processors, file systems, memory disk I/O, NFS, Java™, and communications.
- Operating System and Device Management from IBM provides users and system administrators with complete information that can affect your selection of options when performing such tasks as backing up and restoring the system, managing physical and logical storage, and sizing appropriate paging space.
- IBM Redbooks: The AIX 5L Differences Guide Version 5.3 Edition focuses on the differences introduced in AIX 5L Version 5.3 when compared to AIX 5L Version 5.2.
- Check out other articles and tutorials written by Ken Milberg:
- AIX and UNIX: The AIX and UNIX developerWorks zone provides a wealth of information relating to all aspects of AIX systems administration and expanding your UNIX skills.
- Search the AIX and UNIX library by topic:
- IBM trial software: Build your next development project with software for download directly from developerWorks.