Higher kernel Memory utilization in 8.97 Tools

[riko44]

We have noticed somewhat higher memory utilization on the enterprise server after upgrading our test environments from tools release 8.95 to the tools release 8.97.1.3

Immediately after restarting E1 services (no users connected) each call object kernel takes about 70MB in the 8.97.1.3 vs. about 20MB in the 8.95. Obviously Kernel size growth as more users connect to the kernels in both cases. Nevertheless 8.97 CO kernels seem to be taking on average 20-50% more memory vs 8.95 kernels with similar number of users based on my observations. In fact this seems to be applicable to almost any type of kernel, not just call object kernels.

Most of the enterprise servers have been directly upgraded from one Tools release to another, so kernel settings in Jde.ini between tools seem to be the same. In both cases multi-thread kernels are enabled.

I am wondering if this a normal situation considering some major differences between TR 8.95 and 8.97 or this is something specific to our platform (AIX) and/or setup.
Please note that we are on 8.11SP1, not on 8.12.

Here is an example of our CO kernel definition (the same in both tools)

[JDENET_KERNEL_DEF6]
krnlName=CALL OBJECT KERNEL
dispatchDLLName=libxmlcallobj.so
dispatchDLLFunction=XMLCallObjectDispatch
maxNumberOfProcesses=15
#numberOfAutoStartProcesses=0
numberOfAutoStartProcesses=1
singleThreadedMode=N
ThreadPoolSize=30
ThreadPoolSizeIncrement=10
useBufferLogging=N

Any input is highly appreciated

Thanks

E811SP1, AIX5.3 , Oracle 9.2.0.6

PS Additional Question - What is nowadays considered as a best practice regarding ratio between # of concurrent users and # of call object kernels for 8.97.

 

[FrankJimenez]

We are on AIX 5.3. We upgraded from XE to 8.12 a year ago and upgraded to 8.97.1.3 several months ago. And yes there were significant additional memory requirements. Not apples to apples I know but to get the same level of performance, we had to almost double system ram. We're having problems with kernel memory usage also but from what I understand, best practice is a dozen users per kernel. They will load much higher than that and seem to work ok up to a point.

This from the enterprise server:

[JDENET_KERNEL_DEF6]
krnlName=CALL OBJECT KERNEL
dispatchDLLName=libxmlcallobj.so
dispatchDLLFunction=XMLCallObjectDispatch
maxNumberOfProcesses=38
numberOfAutoStartProcesses=38
singleThreadedMode=N
ThreadPoolSize=30
ThreadPoolSizeIncrement=10

Hope this helps.

Thanks,

Frank

 

[JEMILLER]

Starting from 8.96 tools a number of kernel types load an in-process JVM. (The Java Virtual Machine is loaded into the JDENET_K process. You will not see a separate JVM process). I believe that Call Object kernels do this.

Looking at your JDENET_KERNEL_DEF6 you have multi-threading enable. Multiple threads within the call object kernels in 8.97 is sure increase memory usage somewhat.

I don't have numbers to back these comments up. As you have observed this across all kernel types I would guess that it is the addition of the in-process JVM that has increases kernel memory footprint across the board.

 

[riko44]

Thanks a lot! I am curious which tools Release did you upgrade your 8.12 from several month ago? I am talking about upgrade within 8.12, not Xe->8.12.

[ QUOTE ]
We are on AIX 5.3. We upgraded from XE to 8.12 a year ago and upgraded to 8.97.1.3 several months ago.

[/ QUOTE ]

 

[cmanderson]

[ QUOTE ]
Looking at your JDENET_KERNEL_DEF6 you have multi-threading enable. Multiple threads within the call object kernels in 8.97 is sure increase memory usage somewhat.


[/ QUOTE ]

Not at first - as root you can run 'svmon' and see that, with multithreaded kernels, the memory consumed per process at startup is about the same as in singlethreaded mode.

As threads are launched, sure, it should increase the amount of memory consumed by the process. In theory multiple threads should take less system memory than multiple kernel processes in single threaded mode, but I have such a small installation of 8.11 SP1 on AIX 5.3 that is it hard for me to say one way or the other without load testing.

I totally agree with you about the JVM enhancements. In all I am much more satisfied with 8.97 over the previous 8.96 on similar System p hardware.

-Charles