First BSFN call performance

craig_welton

craig_welton

Legendary Poster
Hello all,

Having a performance issue I was hoping to get some insight on.

E900
8.98.3.4
WebSphere 6.1 (Win32)
iSeries enterprise server

Users complaining about long waits when starting their first application. Tracing the JAS logs shows the first BSFN called is taking over 7 seconds to complete. It doesn't matter what BSFN, just the initial connection to a call object kernel. Once that is made, everything flies.

Below is the Call object kernel config.

[JDENET_KERNEL_DEF6]
krnlName=CALL OBJECT KERNEL
dispatchDLLName=XMLCALLOBJ
dispatchDLLFunction=XMLCallObjectDispatch
maxNumberOfProcesses=50
numberOfAutoStartProcesses=0
singleThreadedMode=N
ThreadPoolSize=20
ThreadPoolSizeIncrement=5

Any hints on where to look?

thanks,
Craig
 
Most likely there is more going on but I'd recommend auto-starting some of those Call Object Kernels. After testing that I'd enable debugging and then run the logs through performance management workbench.
 
Thanks Paul,

Currently, there are 32 call object kernels running. I'm going to work with logging on the enterprise server. A little tricky since their clocks aren't in synch. I'm wondering if it's a websphere config issue?
 
Craig,

We have a similar situation, when user's first log on. I think we had advice that data was being cached to use during the session. But 7 seconds sounds a bit long.
 
Hi,

I'd set the number of autostartprocesses to 10, so
users don't have to wait for them to be initialized.
 
The 7 second delay is the time it takes to "initialize" a user session with the call object kernel. A few months ago, a setting was changed in the LocalLogicCatalog.xml file on the JAS server. The BSFN LeftJustifyUDCValue was marked to run locally. This was supposed to speed up the login process. Well, it did, only to push the delay to the first BSFN call from within an app.

Fair enough. Question to iSeries folks ... is 7 seconds a norm for initializing a user session (getting all environment data, caching OMAP etc.)?

thanks,
Craig
 
Craig,
Depending on the vast multitude of factors involved in iSeries performance, it could be "normal". I have a client on an undersized box that experiences 7 seconds like you.

-Ethan
 
Craig, I am currently looking at the same thing.
The 'Call Objects Stats' in Server Manager shows this is not a BSFN issue, but [init-remote-env], which is not a BSFN.
In our case, over 26 days the cumulative time spent initialising the session on the kernel has been (per JVM) between 2 to 3 hours! Put in this way, it is a real performance loss, and unacceptable.
On the BSFN side, F4311EndDoc (and its associated bsfns) are taking a long cumulative time also - around 1 hour 10 minutes over the same 26 day period.
I have logged this with Oracle support but unfortunately no solution as yet. cheers
 
Back
Top