Best way to start developing using parallel jobs

Mike3000 · Post by **Mike3000** » Mon Mar 26, 2007 10:13 am

I have a good experience of doing development using Server jobs, but
now client wants to use DS Parallel Jobs.

Could you tell me please what is the best approach to make a smooth
transition to the DS Parallel Edition from DS Server Edition? and in
your opinion how difficult/time consuming is it?

DSguru2B · Post by **DSguru2B** » Mon Mar 26, 2007 10:27 am

Welcome Aboard

My advice, go for Server To Parallel transition classes. Ray's class is going on in Dallas as we speak.

ravibabu · Post by **ravibabu** » Mon Mar 26, 2007 11:05 am

Hi

Please try to avoid Xrm and if is there any datatype modification and datatye conversion use the MODIFY stage.

and soon many.....

karthegx · Post by **karthegx** » Mon Mar 26, 2007 11:10 am

ya that's correct use as many diffferent stages avilable in parallel jobs ,reduce most of the xrm stages . and use Remove Duplicates stage for removing duplicates if u have any.........

DSguru2B · Post by **DSguru2B** » Mon Mar 26, 2007 11:13 am

No need to avoid the transformer stage as its performance has been greatly increased in 7.x versions. No doubt, modify stage still leads the two, but transformer is faster than filter stage and a few other stages. IBM released this information sometime back.

ravibabu · Post by **ravibabu** » Mon Mar 26, 2007 11:22 am

Yes ,You may worng sir...Bcz Xrm is generate the c++ code and this stage consumed the large memory compary to other stages.
Other stages generate the hash file entrys.That is the resone to avoid the Xrm stage. Xrm stage want so many section leaders and players.
But other stages not like that.....if i am worng correct me...

Mike3000 · Post by **Mike3000** » Mon Mar 26, 2007 11:51 am

Thank you Guys for your answers.

So, i still can use the old functionality and as i go ,
i can gradually add a new functionality of Parallel jobs.

Could you suggest any books about Parallel Jobs or may be a
good link? It would be greatly appreciated.

DSguru2B · Post by **DSguru2B** » Mon Mar 26, 2007 11:59 am

True. But my comment was w.r.t performance.
Memory demands follow simple economics. Demand and supply. I have never had memory leaks with a transformer stage. I have had memory violations while using custom px routine but that was my own faulty memory management in the C code. Other than that, the transformer jobs run like a breeze.

ravibabu · Post by **ravibabu** » Mon Mar 26, 2007 12:13 pm

Thanks for you reply.

Any one please send the best online guide for PX and Server.

That could be helpful to all.

ray.wurlod · Post by **ray.wurlod** » Mon Mar 26, 2007 5:29 pm

There are no online guides. Only the manuals that come with the product.

ag_ram · Post by **ag_ram** » Wed Jul 04, 2007 2:29 am

DSguru2B wrote:No need to avoid the transformer stage as its performance has been greatly increased in 7.x versions. No doubt, modify stage still leads the two, but transformer is faster than filter stage and a few other stages. IBM released this information sometime back.

Hi DSGuru

Can you share the link where IBM document is available that speaks about comparison of Filter and Transformer (transformer is faster than filter stage and a few other stages. IBM released this information sometime back)
I tried to search IBM and Ascential Developer Net (IBM Forum) , i did not manage to get the same.

ray.wurlod · Post by **ray.wurlod** » Wed Jul 04, 2007 1:01 pm

The information that IBM released "some time back" was for version 7.0 and earlier and has been overtaken by events. Now they're advising (in their training classes) that Transformer should be perferred to Filter, on the grounds that it's compiled and Filter is "interpreted". I believe the justification is spurious, because the filter operator is an instance of an object for which the class is written in C++, so there ought to be little to choose between them. But I have no benchmarks to support that belief.