You are on page 1of 1

4.

7 Parallel/Distributed Consistency Achievement 113

0 otherwise.

Guesgen calls the convergence mechanism graceful degradation. Each v-node out-
puts the signatures that it stores to its connected c-nodes. When a c-node is on, it
takes the intersection of the (two) input sets of signatures and outputs the result to
the v-nodes connected to it; an empty set is output if it is off. Therefore, input to
each v-node is sets of signatures from the c-nodes. These inputs will be combined
using the logic shown in Figure 4.4, with AND operations replaced by set union
operations, and OR operations replaced by set intersection operations. This com-
bined result is intersected with the value currently stored in the v-node. The input,
output and the values of each node are summarized in Figure 4.5.

For the c-node for (<x, a> <y, b>):

input: v(<x, a>) ∩ v(<y, b>)

value (static): 1 if (<x, a> <y, b>) satisfies Cx,y ;


0 otherwise

output: v(<x, a>) ∩ v(<y, b>) if c( <x, a> <y, b>) = 1;


{} otherwise

For the v-node for <x, a>:

input: ∩ ∪ output of the c-node for (<x, a> <y, b>)


y ∈ Z b ∈ Dy

value: current value ∩ input


output: current value (i.e. v(<x, a>))

Figure 4.5 Summary of the input, output and values of the nodes in
Guesgen’s network (Z = set of variables, Dy = domain of y)

The network will always converge because the number of signatures stored in the v-
nodes is finite and nonincreasing. After the network has converged, a solution W to
the CSP is a compound label such that:

You might also like