IBM Mainframe Forum

Hi to everyone,

I believe this is a repeated question, but i couldnt find any answer to my specfic question.
The thing is that i am trying to replace the Easytrieve by a Syncsort because syncsort's performance supposed to be better.

This Eastrieve perform a matching between 2 files and write the output with the not matched records adding a date to the output layout. Records lenght are equals between inputs and output.

I am replacing the easytrieve by the following :

JOINKEYS FILE=F1,FIELDS=(1,4,BI,A),SORTED
JOINKEYS FILE=F2,FIELDS=(1,4,BI,A),SORTED
JOIN UNPAIRED,F2,ONLY
REFORMAT FIELDS=(F2:1,100)
SORT FIELDS=COPY
OUTREC BUILD=(1,15,C'20120329',77X)

I've run a test with 50 million records (representing the 15% of the records currently procesed in prod)

Syncsort : CPU TIME= 1.82 ----- TOTAL ELAPSED TIME= 31.74 ----- JOB COST 8.67
Easytrieve : CPU TIME= 2.02 ----- TOTAL ELAPSED TIME= 12.09 ----- JOB COST 9.79

I am trying to analyze the results with TRITUNE but i am not familiar with it, so....What else should I consider to determine which has better Performance ?


Thanks a lot for help
Sergio

OK, couple of things first with your SORT cards. You can add NOSEQCHK (or however it is spelled, you can CHecK it) after both of the SORTED on your JOINKEYS. This will turn of the sequence checking as well (at least it does in DFSORT, I don't have Syncsort documentation).

Also, you are extracting 100 bytes but blatting the final 77 of them to space. Not something you want to do 300 million times if you can avoid it. Since your positions 1-15 are staying where they are, OVERLAY with an appropriate column number for your date will be better.

Elapsed time is always a tricky thing, as it relies on things external to your run (other stuff using resources). Try to run them consecutively but at around the same time. If you consistently get longer elapsed from Syncsort, you'll have to do some more investigation, but I can't think that you will.

Now, how about showing the Easytrieve as well?

i've checked NOSEQCHK and search for something similar, but i couldn't find what you mention in the syncsort documentation , anyways i redefined the sort card considering what you mention about the overlay , but i had to keep the space formatting at the end of the record.

I run again the syncsort with 50 million records ,and the performance improved:

JOB COST CPU TIME TOTAL ELAPSED TIME
7.23 1.53 6.55


this is the new sort card


JOINKEYS FILE=F1,FIELDS=(1,4,BI,A),SORTED
JOINKEYS FILE=F2,FIELDS=(1,4,BI,A),SORTED
JOIN UNPAIRED,F2,ONLY
REFORMAT FIELDS=(F2:1,100)
SORT FIELDS=COPY
OUTREC OVERLAY=(16:C'20120328',24:77X)
/*

I forgot to mention that i run the syncsort and easytrieve one after the other.


Regards
Sergio

OK. You'll have to experiment yourself with this as I don't have Syncsort documentation.

Try removing the REFORMAT. If you get the wrong results, change the REFORMAT to (F2:1,15).

Strictly you don't need the 24:, because 24 is the next column anyway (check for your product), but I don't think that'll have any impact on performance.

What is the thinking behind 77% of the record being space? If you have performance issues with these files, the "necessity" for such a thing would be something to review.

A couple of other things. You haven't shown the Easytrieve.

It would be useful for you to include a comparison of the EXCPs.

How have you selected the 50million records? Is this a representative sample (of the percentages of match to the different types of non-match - match being your criteria)? If not representative, you could get bitten when you run with full-size files.