Re: Suggestion for improving loop performance

On 19-May-2015 13:49 -0500, Singh, Ashwani (Lake Mary) wrote:

On 19-May-2015 13:17 -0500, Singh, Ashwani (Lake Mary) wrote:

We are trying to process a 69 million records file which is taking
5 hours plus to load into another file. (We have Block *Yes
specified for the input file (with 69M records) We need to bring
the processing down for sure..

We are trying for the multithread as well..

There are two crucial loops that run in this single program and
causing trouble.

The first loop takes round about .6ms/rec <ed: corrected from
seconds/rec> processed the other one .4ms <ed: corrected from
seconds>.

Our file is a flat file and is | delimited ..

Loop 1:

D ARRAY1 S 5 0 DIM(999)
C '|' SCAN DATA ARRAY1
C Eval X= 1
C Eval A= 0
C Dow ARRAY1(X)> 0
C Eval A+=1
C Eval X+=1
C ENDDO

Loop2:

C DOW A < #Pipesreq
C* <ed: DOW is for the> (Number of pipes actually required)
C Eval Fileds = %subst(fileds:1:%checkr(' ':Fileds)) + '|'
C eval A+=1
C ENDDO

Any suggestion will be highly appreciated..

<<SNIP>>
This is what I am trying to do -

We have one control record running into multiple records of flat file
i.e.

X:01 A|B|C
X:02 D|E|F
X:03 G|H|I

X:01 1|2|3|4|5
X:02 ......


And I know that this record X:01 should contain 5 fields but have 2
missing and rest all are fine.

The output becomes like this

X: A|B|C||D|E|E|G|H|I

Perhaps instead, output should be?:

X: A|B|C|||D|E|F|G|H|I

This whole process is repeated like this...

<<SNIP>>

Suggestion: Do not interrogate how much work is to be done before actually doing the work; much time can be saved, because the pre-processing that inherently requires two passes is eliminated, so the data need be processed just once.

In other words: drop the upfront SCAN, and just start parsing the data. When the EOR occurs before all elements are accounted, then instead of tacking on the missing delimiters, know that the missing elements are representing effectively a NULL value, and process them as such.

FWiW: If the problem scenario is described explicitly as an example with the setup creating the object(s) and input of some sample data [a rendition of actual data or facsimile], then the textual descriptions of the problem such as "this record X:01" are either eliminated as confusing as they are, or they take on some actual meaning. A reviewer might see an alternate approach for the problem scenario, rather than concentrating on the minutiae of two specific loops... which as I allude, may not be worth keeping at all.