BW Process Chain Design

[Former Member]

Queries on BW process chain design(especially about meta-chains,which process types to use etc) regularly appear in the BI forums.This blog attempts to give some know-how on this topic.(Information is applicable till BW ver 3.5)

• For Data Load from Source System to Infocube
• 
  
  Steps..
  
  Start Variant-->Delete Indexes on the Cube-->Infopackage-->Generate Indexes on Cube-->...
  
  Index creation step can be followed by steps for constructing DB Statistics,deletion of overlapping requests,roll up of filled aggregates,compression etc(in that order).Links:1,2.
  
  If complete data target contents are to be deleted before each data load(used in case of full updates to data target),use process 'complete deletion of data target contents'.
  
  Start Variant-->Complete Deletion of Data Target Contents-->Delete Indexes on the Cube-->Infopackage-->Generate Indexes on Cube-->...
  
  
  
  
• For Data load from Source System to ODS and then to further Data Targets
• 
  
  Steps..
  
  Start Variant-->Infopackage to Load to ODS-->ODS Activation-->Delete Indexes on Target Cubes-->
  Further Update from ODS to Targets-->Generate Indexes on Target Cubes-->...
  
  If multiple infopackages load to the same ODS,you do not need to place multiple ODS activation steps.All unactivated requests loaded to the ODS will get activated in a single activation step.Link:3.
  
  Important Note:The checkboxes for 'Activate ODS object data automatically' and 'update data targets from ODS Object automatically'
  in ODS->Change screen hold no significance when ODS is loaded in process chains.Irrespective of the checkboxes being ticked or not,steps for both these activities have to be explicitly placed in the chain(as shown in above steps).
  
  
  
  Similar concept applies to infopackages that use setting 'only PSA,update subsequently to data target' in the processing tab.Placing only this infopackage in the chain will bring data only till the PSA.
  Data will goto data target only when you use the process type 'read PSA and update Data Target'(place it as the immediate step after the infopackage).Links:4,5.
  
  
  
  
  
  
• For Master Data Loads
• 
  
  Steps..
  
  Start Variant-->Infopackages to Load Master Data...--->AND process-->Hier/Attr Change Run
  
  In the variant for Hier/Attr change run,specify all the MD infoobjects that are getting loaded in that chain.
  
  
  
  AND process is a 'collector' process.It triggers the subsequent process(es) when ALL of the steps which link to it(or precede it),are successful.
  To explain,in above example:the hier/attr change run should run only after all master data infopackages in the chain have run successfully.So AND process is used.
  
  Other collector processes are OR and EXOR(exclusive-OR) process.
  OR process triggers next process when any of its preceding steps runs successfully.
  EXOR process triggers next process only once(as opposed to the OR process) when any of its preceding steps first runs successfully.
  
  
  
  
• Running Chains within Chains(Meta-Chain concept)
• 
  
  When a process chain contains other process chains in its steps,it is known as a 'meta' chain.And the chains contained in it are called as 'local' chains.
  
  We use this concept when we want the execution of a chain to be dependent on the successful execution of one or more chains.While creating a meta-chain,use process type 'Local Process Chain' to insert a local chain as one of its steps.
  
  
  
  Note that for 'local' chains,start variant should have the setting 'start using meta chain or API' and not as 'direct scheduling'.
  
  
  
  For more information please see this link(help.sap for BW 3.5 'process chains' section)