Process flow/map performance issues

Question

We have some issues with our OWB-based application and we're looking to find out if there are different ways we could be using the tool, or features/options we've missed.
We are trying to maintain a near real time feed of data from a front end system into our warehouse which was built using OWB 10.2.0.3 over a 10.2.0.4 database. The bulk of the application consists of OWB maps with a few hand-written PL/SQL objects, all executed in a series of hierachical OWB process flows. Maps/transformations are executed either sequentially or in parallel where the referential integrity of the model allows.
The problem is that we have around 150 tables in the datamart which could potentially require updating on each refresh cycle, although in practice only a few tables have any activity on a typical refresh cycle. The cycle consists of loading data into a set of staging tables, and from there the data is transformed into the main schema, often with multiple maps per target table.
On every cycle we run hundreds of maps, the vast majority of which process zero rows. Each map runs quickly and efficiently in its own right but collectively they add up to a 5 - 10 min cycle even if there is no data to process.
There are 2 avenues which we'd like to explore and would be grateful if anyone could provide any pointers/suggestions :-
1) It appears that each map opens and closes its own database session when it executes. I presume this was done because a single process flow could be constructed with maps executing in different target schemas, but we know that's not the case for us. We'd like to know if there is anyway to configure the database connection at a higher level (eg. process flow) so it opens a connection once and executes each of the maps (database packages) in that one session.
Our DBAs are experimenting with 'shared server' settings at a database level which may help to some degree but won't be the whole story.
2) Another option is simply to run less maps eg. load the staging area as now, collate stats on which staging tables contain new data, and then apply some logic such that subsequent maps only execute if the relevant staging table(s) contain(s) some new data, otherwise bypass that map.
We tried experimenting with the 'Pre Mapping Process' operator, but essentially that just generates another function call from the map package, so we still have the overhead of opening a database session for each map to run the package. Minimal gain.
We thought about adding a function call in the process flow before each map and then branching to either execute/bypass the map as approriate, but the function call still requires opening/closing of a database session each time so, once again, minimal gain.
What we really want is some way for a map or process flow to check without logging onto the database repeatedly.
Any ideas on the above, or other potential solutions anyone could suggest, would be greatly appreciated.

Answer 1

Hi,
Please see if these documents help.
Note: 554635.1 - Create Accounting Process Performs Poorly When 100K + Distributions are Passed for an Event
Note: 954273.1 - Multiple Create Accounting Requests Result In Poor Performance For Online Accruals
Note: 763500.1 - R12: Performance Issue with Create Accounting
Note: 733637.1 - R12:Performance Issue When Running Accounting Program Xlaaccup
Note: 781311.1 - Create Accounting Process Taking A Long Time To Complete After Appying Critical Patches
Note: 557869.1 - EBS: R12 Oracle Financials Critical Patches
Regards,
Hussein