Subject: RE: EXEC CICS Syncpoints
From: Edward M. Martin (EMartin@aultman.com)
Date: Thu Jun 05 2008 - 10:00:07 EDT
Hello Martin,
I think that I explained this wrong.
ZMID is transaction 1 running.
Read ZMIDAMD record
Write ZMISSMD
Do a SyncPoint every 10.
Was determine that
we would back out 10
Delete ZMIDAMD record
SyncPoint
every time
HUBM is transaction 2 running
Read ZMISSMD
Send via TCP to EGATE server
sends record to multiple
systems.
Delete ZMISSMD.
SyncPoint.
every time
The problem is these two files are ZMISSMD and ZMIDAMD have
normal amounts of I/O requests, but tremendous EXCPs
From TMON
File ID/Type/VOLSER I/O Requests EXCP Count Avg
Response
<----216,944-----> <----216,944----->
<------.200------>
ZMSIIMD V DSK7BD 35,845 48778755
3.250
ZMIDAMD was just as high when ZMID was running.
I was thinking that the SYNCPOINT force the physical write of
the buffers to each file causing an I/O for each READ/WRITE and a
separate EXCP for each SYNCPOINT, although the numbers do not work out.
I just can't justify a syncpoint for each record in ZMID.
Ed Martin
330-588-4723
ext 40441
-----Original Message-----
From: owner-vse-l@Lehigh.EDU [mailto:owner-vse-l@Lehigh.EDU] On Behalf
Of Martin T2..
Sent: Thursday, June 05, 2008 1:30 AM
To: VSE Discussion List
Subject: Re: EXEC CICS Syncpoints
Hello Ed,
There is still something missing (at least my feeling says so).
Let me repeat what I did understand.
File is created in batch- okay
loop until (maybe no more records)
read ZMIDAMD
write ZMISSMD
loop 10(or 9) times
SYNCPOINT
delete record from ZMIDAMD
end-lopp
SYNCPOINT
end-loop
Why are 9 (or 8) records deleted without being acted upon?
or is the read in fact a read against multiple records?
or can the records be deleted by a GENERIC key?
The SYNCPOINT in the inner loop causes on first the changes against
ZMISSMD being commited- the further SYNCPOINT cause the delete to be
committed. It could be that CI-locks against the CI- where a record was
deleted triggered (while testing) the extra SYNCPOINT.
If the above is true (testing showed...) I would recheck my design and
have the info of the extra 9 (or 8) record in the same record or give
them at least the same key so one can use GENERIC and only a single
DELETE. Remember ACID!
In view of further growth I would (after above change implemented) do
the READ with UPDATE then the DELETE or (in case of multiple records)
ENQUE on some argument... this would enable multiple processes to work
on incoming data.
--
Martin
--
XML2PDF - the way to get all features of PDF into your documents
on mainframe or PC systems; more at http://www.pi-sysprog.de
This archive was generated by hypermail 2b25 : Sun Jul 27 2008 - 05:20:06 EDT