Cost-Based Optimization of Integration Flows - Datenbanken ...

6.4 Optimization Techniques
Algorithm 6.4 Update via Partial PlanOptTree Replacement (A-PPR)
Require: set of new partial PlanOptTrees PPOT, invalid optimality conditions OC
1: for all oc ∈ OC do // remove invalid optimality conditions
2: oc.op1.ocnodes.remove(oc)
3: if oc.op1.|ocnodes| = 0 and oc.op1.|csnodes| = 0 then
4: rremoveNodes(oc.op1) // recursive bottom-up
5: oc.op2.ocnodes.remove(oc)
6: if oc.op2.|ocnodes| = 0 and oc.op2.|csnodes| = 0 then
7: rremoveNodes(oc.op2) // recursive bottom-up
8: clearStrata12()
9: for all ppot ∈ PPOT do // merge new partial PlanOptTrees
10: mergePPOT(root, ppot) // see A-IPC lines 11-22
starting from the root with the same concept of removing nodes without any children.
Finally, we apply the merge algorithm (line 10) from Subsection 6.2.2.
With the aim of reuse, we could index plans and PlanOptTrees created over time by
their optimality constraints. This could avoid redundant directed re-optimization and
merging PlanOptTrees but we would still need to copy statistics. Due to the risk of (1)
maintenance overhead (e.g., plans, PlanOptTrees), (2) a potentially large search space, as
well as (3) low remaining optimization potential, we do not reuse plans. However, future
work might investigate this by combining on-demand re-optimization with (progressive)
parametric query optimization (PPQO) [BBD09] by iteratively creating possible plans of
the search space according to the optimality conditions and subsequently, reusing already
created plans.
To summarize, we have shown how to use the PlanOptTree for directed re-optimization
and how a PlanOptTree can be incrementally updated after successful re-optimization.
All algorithms presented rely on the extension of the optimizer interface by returning
partial PlanOptTrees or by directly rearranging the referenced PlanOptTree. Either way,
existing optimization techniques require modifications. In the following, we will explain
these modifications using selected optimization techniques from previous chapters.
6.4 Optimization Techniques
In order to illustrate the applicability of the PlanOptTree and the necessary modifications
of optimization techniques, we use examples to show how their optimality conditions can
be expressed with our approach. First, we describe the on-demand re-optimization for
common rewriting techniques (see Chapter 3). Second, we show how the concept of ondemand
re-optimization can be applied for cost-based vectorization (see Chapter 4) and
multi-flow optimization (see Chapter 5) as well.
6.4.1 Control-Flow- and Data-Flow-Oriented Techniques
For control-flow- and data-flow-oriented rewriting techniques, their already presented optimality
conditions can be reused as they are. In this subsection, we discuss the on-demand
re-optimization for the common data-flow-oriented example optimization techniques join
enumeration, eager group-by, and set operations with distinctness.
181