DSXchange

Hi,

Tell me, where it is preferred to have IPC stage and where it is preferred to have Link Partioner & Collector stage.

Welcome aboard. :D

They are unrelated. IPC implements and makes visible pipeline parallelism in server jobs, while the Link Partitioner/Collector stages implement and make visible partition parallelism.

For more information read Chapter 2 of Parallel Job Developer's Guide

If two active stage are connected to gether, the process are clubbed together. If an IPC is used in mid of it, the both process are mutually exclusive.
If you need parallalism, you can use Link partitionor. If you have 64links out of it, you input record set will be processed parallely in all the 64 link, each record in each link.

If two active stages are linked together, and inter-process row buffering is enabled, then the two active stages will operate in separate processes whether or not there is an IPC stage placed on the link. The IPC stage simply makes the process boundary visible in the job design, and gives a mechanism for changing the buffer size and timeout specifically for that link.

IPC would be a key through with the link level row buffering parameters (buffer size and timeout )can be altered or changed. (The value, if set in project level in row buffering)