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

6 On-Demand Re-Optimization
o 3 o 4
o 2 o 3
(a) Partial POT 1
|ds in1| |ds out1| |ds in1| |ds out1| |ds in1| |ds out1| |ds in1| |ds out1|
sel(o2)
sel(o3)
sel(o3)
sel(o4)
≤
≤
o 2 o 3
|ds in1| |ds out1| |ds in1| |ds out1|
sel(o2)
sel(o3)
≤
≤
(b) Partial POT 2 (c) Merged POT
Figure 6.6: Merging Partial PlanOptTrees
o 4
|ds in1| |ds out1|
sel(o4)
invoke this algorithm (line 4), where all subcalls can access the existing PlanOptTree
root. Otherwise, we apply all optimization techniques and obtain the resulting partial
PlanOptTree for the atomic operator (line 6). If no PlanOptTree exists so far, the first
partial PlanOptTree is used as root; otherwise, we merge the partial PlanOptTree with the
existing root (lines 7-22). When merging, we first clear the MEMO structure (line 11). Then,
we check for the existence of operators as well as statistic nodes, and we add the new nodes
if required. At the level of complex statistic nodes, we invoke the modifyDanglingRefs
in order to change the references of complex statistics and optimality conditions to the
existing PlanOptTree. We use the MEMO structure in order to mark already processed
paths of the new partial PlanOptTree. As a result, we guarantee a worst-case number
Algorithm 6.1 Initial PlanOptTree Creation (A-IPC)
Require: operator op, global variable root (initially set to NULL)
1: o ← op.getSequenceOfOperators()
2: for i ← 1 to |o| do // for each operator o i
3: if type(o i ) ∈ (Plan, Switch, Fork, Iteration, Undefined) then // complex
4: o i ← A-IPC(o i )
5: else // atomic
6: ppot ← getPartialOptTree(o i )
7: if root = NULL then
8: root ← ppot
9: else
10: // merge partial PlanOptTrees
11: clear memo
12: for all on ∈ ppot.onodes do // for each ONode on
13: if root.containsONode(on.nid) then
14: eon ← root.getOperator(on.nid)
15: for all sn ∈ on.snodes do // for each SNode sn
16: if eon.containsSNode(sn.type) then
17: eson ← eon.getSNode(sn.type)
18: modifyDanglingRefs(eon, eson, on, sn)
19: else
20: eon.snodes.add(sn)
21: else
22: root.onodes.add(on) // add operator subtree
23: return root
174