Tuning querying and indirection

Question

We have a legacy design where everything is highly normalized, even in the object model, and uses indirection for all contained parts of an object. This works well for most of the needs in the application. However, we are trying to perform a calculation that starts with a result set of root objects and navigates through the owned parts of the root object (1 to many relationships AND 1 to 1 relationships) to calculate an aggregate value. Depending on the business rules and the state of the root object, different parts of the root object are required when doing the calculation for the aggregated sum (hence the difficulty of using partial object reads via ReportQuery).
Is there any way to tune a query to trigger indirection on an object for all its owned parts (i.e. load them into memory immediately)? Loading everything into memory when we query for the root objects would be fine. Delayed querying would be more expensive, we believe. We also don't want to change the descriptors globally because indirection is appropriate for many of the other scenarios in the code.
Other recommendations for doing the calculation without the overhead of indirection and its multitude of queries would also be appreciated.

Answer 1

TopLink provide its "batch reading" feature to optmized the loading of a large graph of objects. The main cost in traversing a large graph of objects is that you get a lot of small queries. Refer to the documentation section on batch reading and the query API addBatchReadAttribute.
If you have A->1-m->B->1-1->C and initially read n As that each have m Bs, normally reading will require 1+n+n*m queries (if n was 100 and m 10 = 1101). With batch reading this can be converted to 3 queries giving a huge performance benifit.
To force indirection to fire to a certain level, you can either walk your object yourself, or call the TopLink API, copyObject(level) to force instantiation.