Of course, we all watched or maybe you were at location in Munich, July 8, 2025 to be at the introduction of the IBM Power11.
I also got an invitation form Chris Gibson for the launch events in Sydney 5 August, Melbourne 12 August and Brisbane 12 August. I really wish I could be there!
This number 11 is a true milestone again in the Power era, and with very important features that guarantees zero downtime for our applications and databases.
More Important numbers in July
But another number (date) in the month July was very important for AIX and Power.
25 July 28, 2025!
On this date AIX SP1 for AIX 7.3 TL03 was released and became available for us.
For those who are reading my blogs, you know that I really like the new features in AIX 7.3 TL03 and the new technology and improvements that came with it.
But with AIX 7.3 TL3 SP0 not all the improvements let to better performance, at least not for rather small LPAR’s.
Let me explain more what happened with this particular level and LKU (Live Kernel Update).
Development of IBM worked hard to get better blackout numbers for large LPAR’s therefore they made several memory optimizations. This worked out for large memory and huge numbers of filesystems.
Disadvantage was that for rather small LPAR’s the opposite was achieved, I wrote in one of my earlier blogs this year that the blackout time for those LPAR’s raised form a blackout time AIX 7.2 TL05 SP9 = average 12 sec to AIX 7.3 TL03 SP0 to 18 sec.
Those 6 seconds seems not be a lot right?
But what if you have to pay a fee for every second (transactions) your application is not responding?
And what if in some (rare occasions) the blackout time became more that one minute stead of 12 secs.
Well good news last week I did a rerun the same tests again, and I was very pleased to see that the blackout time was back to around the familiar number 11 sec again.
Also the overall process was also much shorter.
See this article posted by Vinod Boddukuri
AIX LKU performance improvements in AIX 73 TL3 SP1
that shows the blackout times for rather big Oracle LPAR.
I liked to see what happens with our own environment, and my test environment are rather small LPARS that, are running an IBM LDAP environment, with underneath DB2 as database and a WebSphere application for management.
This environment is an ideal test environment because I am familiar with the blackout times and overall times on AIX 7.2 and on AIX 7.3 TL03 SP0.
I run the test several times an for me the most important number was the blackout time.
The average of 3 test was just below 12 seconds I was very pleased to see this nice number of 11 again:
On the standard output I says however 13 seconds:
07/30/2025-13:52:06 Blackout Time started.
07/30/2025-13:52:19 Blackout Time end.
But more precise measurement from the alog:
alog -t mobte -o | sed '/^$/d'| tail -1 | cut -d ' ' -f 40
so this test run had an actual backout time of: blackout=11.440255s
Also indeed due to multithreaded optimalisations of the complete LKU process the overall time was indeed reduced to in our case again the magic number 11 to 12 minutes for this small test environment, measured this with the time command real 11m12.720s
lparstat of this environment:
lparstat -m
System configuration: type=Shared mode=Uncapped smt=8 lcpu=8 mem=5120MB psize=1 ent=0.20
More good numbers:
Also the number of real processes that could now properly tested with another new technology LLU (Live Library Update) was now relevant. This because a real LLU test with updated libraries could now be tested after deploy of SP1 and LKU (Live Update).
Since the introduction of LLU with AIX 7.3 TL03 SP0 we could not really test our programs that made use of shared libraries libc.a and libpthreads.a.
See also Chris Gibson and my blog earlier this year:
https://community.ibm.com/community/user/power/blogs/christian-sonnemans1/2025/01/30/testing-aix-73-tl03-new-features-and-lku
But now after update our LPAR form SP0 to SP1 we had several updated executables that do make use of those updated two shared libraries.
For the test I used again the same LPAR’s.
And after a successful upgrade to SP1, of course with LKU, I run my first llu update manually.
See below the results:
# llvupdate -a -n2 -t30
Non-interruptable live library update operation begins in 10 seconds.
llvupdate preview
Validating new module /usr/lib/libpthreads.a(_shr_xpg5.o)
Validating new module /usr/lib/libc.a(_shr.o)
Adding the process init pid 1 to the list
Validating new module /usr/lib/libpthreads.a(_shr_xpg5.o)
New module has additional exports /usr/lib/libpthreads.a(_shr_xpg5.o)
Validating new module /usr/lib/libc.a(_shr.o)
New module has additional exports /usr/lib/libc.a(_shr.o)
Adding the process secldapclntd pid 3080698 to the list
Adding the process nimon pid 3932504 to the list
Adding the process nimsh pid 4194708 to the list
Validating new module /usr/lib/libpthreads.a(_shr_xpg5_64.o)
New module has additional exports /usr/lib/libpthreads.a(_shr_xpg5_64.o)
Validating new module /usr/lib/libc.a(_shr_64.o)
New module has additional exports /usr/lib/libc.a(_shr_64.o)
Adding the process backup_agent_main pid 4325786 to the list
Adding the process ntpd4 pid 4456872 to the list
Adding the process bootstrap_agent_main pid 5046580 to the list
Validating new module /usr/lib/libpthreads.a(_shr.o)
New module has additional exports /usr/lib/libpthreads.a(_shr.o)
Adding the process portmap pid 5308900 to the list
Adding the process iked pid 6160782 to the list
Adding the process biod pid 6554070 to the list
Adding the process aso pid 6619634 to the list
Adding the process srcmstr pid 6881760 to the list
Adding the process netcd pid 7143692 to the list
Adding the process sshd pid 7602450 to the list
Adding the process ksh93 pid 7864828 to the list
Adding the process BESClient pid 8061234 to the list
Adding the process cpsd pid 13107690 to the list
Adding the process syslog-ng pid 14746054 to the list
------------------------------------------------------------
************************** TRY 1 **************************
======== NOTIFY PHASE BEGIN ======== | try 1 | tonotify:18
======== NOTIFY PHASE ENDED ======== | try 1 | suspended:18
...............................................................
======== QUERY PHASE BEGIN ======== | try 1 | suspended:18
======== QUERY PHASE ENDED ======== | try 1 | updated:13
End of try 1/2, number of processes to be retried 0
====================
==== LLU Report ====
14746054 syslog-ng SUCCESS
13107690 cpsd SUCCESS
8061234 BESClient FAILURE
7864828 ksh93 SUCCESS
7602450 sshd SUCCESS
7143692 netcd FAILURE
6881760 srcmstr SUCCESS
6619634 aso FAILURE
6554070 biod SUCCESS
6160782 iked SUCCESS
5308900 portmap SUCCESS
5046580 bootstrap_agent_main FAILURE
4456872 ntpd4 SUCCESS
4325786 backup_agent_main FAILURE
4194708 nimsh SUCCESS
3932504 nimon SUCCESS
3080698 secldapclntd SUCCESS
1 init SUCCESS
==== LLU Report End ====
SUCCESS: 13 NOT UPDATED: 0 FAILURE: 5
===============================================================
LLU operation is completed.
This was not bad at all most failures are executable that we normally stop before executing a LKU. Those failures were form backup software, netcd (dns cache) ILMT.
For those who remember my session last year we stop those processes in stage 2 prep stage, and start those again in post stage 4. Remember this is technology is still under technical preview, but we as customers can help IBM development to fine tune this very nice (new) feature of AIX.
So I strongly recommend that if you have some spare capacity and if you like to invest in your knowledge test with this new technology and your own applications.
And if you like to know all the details and hand on experience join us at the TechXchange 2025 in Orlando. Chris Gibson will provide the labs, and together we have a technology breakout session #1194 about LKU and new features such as LLU.
Thank word:
Thanks to Vinod Kumar Boddukuri for providing us with more technical details about this subject: https://community.ibm.com/community/user/blogs/vinod-kumar-boddukuri/2025/07/25/aix-li
And thanks to Vinod and the IBM team for those big improvements for LKU and the further development of LLU.
Also thanks to Carl Burnett, Chris Gibson and not to forget Nayden Stoyanov for all the support.