Performance problems with JOB submission with mfbsi in PAC

Hi everyone,

I'd like to know if there's any configuration to set the same REGION where a CTRL-M job is submitted and processed in the PAC?

Current situation, random distribution:
CTRL-> SERVER_01 -> MFBSIJCL -> PAC -> SERVER_02
CTRL-> SERVER_02 -> MFBSIJCL -> PAC -> SERVER_05
CTRL-> SERVER_03 -> MFBSIJCL -> PAC -> SERVER_01
CTRL-> SERVER_04 -> MFBSIJCL -> PAC -> SERVER_03
CTRL-> SERVER_05 -> MFBSIJCL -> PAC -> SERVER_04

Desired situation, 1-to-1 distribution:

CTRL-> SERVER_01 -> MFBSIJCL -> PAC -> SERVER_01
CTRL-> SERVER_02 -> MFBSIJCL -> PAC -> SERVER_02
CTRL-> SERVER_03 -> MFBSIJCL -> PAC -> SERVER_02
CTRL-> SERVER_04 -> MFBSIJCL -> PAC -> SERVER_04
CTRL-> SERVER_05 -> MFBSIJCL -> PAC -> SERVER_05

We're experiencing performance issues, where it's taking 10 seconds, 20 seconds, 30 seconds, or even minutes, between:
MFBSIJCL -> PAC -> SERVERS

We've noticed that when it reaches the desired state (1-to-1 distribution), the time is between 1 and 2 seconds.

Settings:
---------------------------
File mfbsi.cfg
AutoEdit
SkelProcessType=MFBSICTM
StartOfWeek=Sunday
JCL-Ext=jcl
REGION=REGION01
ES-SHUTDOWN-JWAIT=CONTINUE
ALERTBACKTIMEOUT=0
STARTUPTIMEOUT=0
RUNTIMEOUT=0
ES_PAC=PACNAME
ES_PAC_ENDPOINT=SORNAME=redis,sentinel.com.br:26385
JOBSTART-LOOKUP-FREQ=1
JOBEND-LOOKUP-FREQ=1
STDNUM=ON
CASAPI
MFBSISEC=MFBSISEC
ES-Server=-rREGION01
ES-JOB-KILL=YES
KEEP-SUBST-JCL=Y

Submission command line:
---------------------------
mfbsijcl32 -r$ES_SERVER -x$TMPDIR/$JOB_NAME SJOBID=JCL  

------------------------------
Daniel Gennari
Tech Arch Manager
Banco Santander (Brasil) S/A
Sao Paulo BR
------------------------------

Hi Daniel,

Have you considered to use the approach of controlling this via JCL initiators or classes, e.g. only define class z on one region, then any job that is defined with that class will only run on the region will that initiator or class set. Does this help to answer your question?

Regards,
Kim

Hi Kim, thanks for responding!!

Yes, we considered this approach, but since we have 90 machines, there wouldn't be a CLASS available for all of them.

The point is, after CTRL-M submits a JOB, the CTRL-M AGENT installed on the machine calls a shell script. Within this script, there's a command line with mfbsijcl32 -xJOBNAME.jcl ....

Then mfbsijcl32 dispatches the JOB to the PAC, and the PAC decides which of the 90 available machines can receive this JOB for processing.

This is where we're experiencing a considerable delay, as the job remains in the Input Queue and, after a while, enters execution.

As mentioned earlier, we noticed that when a job is executed in the region of the same machine where the CTRL-M AGENT dispatched it, this delay doesn't occur.

The request is: would it be possible to somehow force or set all jobs dispatched by the CTRL-M AGENT to first be served in the same JES region of the same machine, only if the JES region of that machine is not active would it then be submitted to any JES region of the PAC.

Hi Daniel,

I recommend you to consider raising a support case to seek further assistance, e.g. to see exactly where the delay is and if that area can be improved, it may need further investigation and diagnostics/tracing requests. Please include MFESDIAGS diagnostics collection, that helps to investigate as a starting point, along with any evidence that you can share regarding where you believe the delay is.

Regards,
Kim