COMP-3 conversion in Datastage - Strategies

[pani]

Hai

We are trying to convert flat file from Mainframe to load DB2 dataware house hosted on Unix. In this process to convert COMP-3 fields we have comeout with following three approaches .

1) Using server Job and CFF Stage . But we are apprehensive of "Performance Issue" , Are we correct

2) Using Parallel job and "BASIC Transformer" . Here also we are apprehensive of performance issue

3) Using Parallel Job and External routine , using object code of C program.

Can anyone suggest better approach than this in terms of performance , if we don't have alternative , which one of the three will be better choice..

Thanks
Pani

[pani]

Any ideas on this highly appreciated. I think this will be vital to all of us to discuss on this . Please share your ideas.

Any comments from DataStage team are also welcome

Thanks
Pani

[ray.wurlod]

Since no-one seems to have an answer, why not research the three methods yourself and share your results?

Don't forget to document the hardware and software configuration and the volume of data processed (and make sure that this is enough to generate statistically meaningful results).

[clshore]

For those of us who do not 'bleed IBM Blue', please elaborate on Comp-3 format.
Over the years, I've converted just about all of them, using everything from assembler to Ksh (don't ask!) to DataStage.
Since I don't use the info very often, I don't try to remember the nuances; I grab an IBM reference when I need details.

Carter

[chulett]

A quick Google turned this up. It even quotes from an IBM reference.

[shawn_ramsey]

Hai

We are trying to convert flat file from Mainframe to load DB2 dataware house hosted on Unix. In this process to convert COMP-3 fields we have comeout with following three approaches .

1) Using server Job and CFF Stage . But we are apprehensive of "Performance Issue" , Are we correct

2) Using Parallel job and "BASIC Transformer" . Here also we are apprehensive of performance issue

3) Using Parallel Job and External routine , using object code of C program.

Can anyone suggest better approach than this in terms of performance , if we don't have alternative , which one of the three will be better choice..

Thanks
Pani

Pani,

We are running Windows so I cannot speak for anything on the Parallel palette but we did find with the Server jobs it was faster to use the CFF stage over running the Comp, EBCDIC, etc. conversions in a transformer stage (even with the splitter combiner set for 2-3 parallel streams). What worked best was for us was the following:

CFF -> IPC -> Xfrm -> Output

I hope this helps a bit.