Round Robin method in Link Partitioner

ravij · Post by **ravij** » Mon Mar 13, 2006 3:40 am

Hi,

How will the Round Robin method in Link Partitioner stage work? I loaded 12 records from src to tgt using 3 XFM stage b/w Link Partitioner n Link collector stages. Src is Seq file n Tgt is DB2. when i load the data it distributed 2 records to 1st XFM and 4 reocs to 2nd XFM and 6 recos to 3rd XFM. How it internally distributs the records in Server Jobs. In Parallel jobs it is explained clearly but no information in server gude.

any assistance would be appreciated.
thanks in advance.

ravij · Post by **ravij** » Mon Mar 13, 2006 3:45 am

Sorry my dear gurus,

My question is not about Round Robin its about Random method.

I am extremely sorry for giving wrong name.

ArndW · Post by **ArndW** » Mon Mar 13, 2006 3:46 am

That is very odd behaviour; since the round-robin will distribute the rows one at a time to each successive link. You should be getting 4,4,4 rows going down each path. I've just done a quick test and that is the behaviour I am seeing. Are you 100% certain that you've specified round-robin in your compiled job?

{Added 5 minutes later}
I just saw that you posted another question with this same job description using hashed partitioning. I'm fairly certain that your test run either used an old compiled image with hashed partitioning or that something similar happened. Round-robin will distribute rows as evenly as possible.

ravij · Post by **ravij** » Mon Mar 13, 2006 3:58 am

Hi ArndW

Thanks for your quick reply.
I am very sorry for giving wrong information. Actually the questin is about Random Method in Link Partitioner.
How will the Random method in Link Partitioner stage work? I loaded 12 records from src to tgt using 3 XFM stage b/w Link Partitioner n Link collector stages. Src is Seq file n Tgt is DB2. when i load the data it distributed 2 records to 1st XFM and 4 reocs to 2nd XFM and 6 recos to 3rd XFM. How it internally distributs the records in Server Jobs. In Parallel jobs it is explained clearly but no information in server gude.

could you give me some details about the same question regarding Random Method.

kumar_s · Post by **kumar_s** » Mon Mar 13, 2006 4:03 am

The name explains.
It distributes randomly (with a sequence of logig applied) which may be unpredictable.
You may not be sure that you can get the same data in the same link if the input data is added with new set of record or deleted with some for the next run.

ArndW · Post by **ArndW** » Mon Mar 13, 2006 4:24 am

The basic concept is the same in server as it is in parallel. The random method will use the system's internal pseudo-random number generator to give a good distribution. A sample of 12 records is too small, if you had 1000 records you would most likely get a relatively even distribution, the higher the number of rows in a sample the better the distribution. I can't recall if you can seed the random number - that would achieve a reproduceable distribution on each run.