Information Technology Reference
In-Depth Information
(11)
N
t
()
+
ω
⋅
N
x
()
n
w
()
n
w
F
ϕ
()
=
⋅
----------------------------------------
-------------
N
,
where
F
ϕ
()
— fitness function value computed by the
i
-th autonomic node;
N
t
i
()
i
and
N
x
()
are respectively the number of nodes, with which node
i
has trust, including
i
itself, or trust in progress relation per advertised workflow,
N
— total amount of
community nodes the
i
-th node is aware of, including
i
itse
lf;
ωω
i
— relative
importance of trust as compared to trust in progress;
n
w
()
n
,
— the number of work-
flows advertised by the
i
-th node and the total number of community workflows the
i
-
th node is aware of, including
i
-th own workflows. Note, that for the example in sec-
tion 3.1 in the state reflected by (10) nodes would have computed their fitness as in
the second column of Table 1.
This sample calculation demonstrates an interesting property, node fitness is a
function of community awareness. Just two nodes having established trust for all
advertised workflows will have maximum fitness 0,50. The (11) shows that the fittest
node should have some community information beyond current communication sce-
nario; this information can be treated as community context, as degrees of freedom
that can be used for adaptation. For example, as shown in (10) and (11) nodes
B
and
C
are fit not only because they have more trust relationships with other nodes than node
A
but also because they have more residuals in (10).
Community fitness (12) is a generalisation of (11). Maximal community fitness is
always higher than that of a single node. The function (12) for a community (
C
o
) is
non-linear (see table below) with regard to contributions of community members.
Thus we propose that nodes compute their weighted fitness as (13); this computation
for the situation (10) is presented in Table 1, last column.
,
∈
[
01
,
]
[
N
t
()
+
ω
⋅
N
x
()
]
-
N
(12)
C
o
F
ϕ
()
=
--------------------------------------------------------------
n
w
()
i
;
F
ϕ
∈
[
01
,
]
;
i
=
1
N
,
(13)
C
~
F
()
=
1
-
F
⋅
F
()
C
~
ϕ
,
w
ϕ
ϕ
;
i
⊄
Table 1.
Weighted fitness for situation (10)
Node
Node
Fitness
Community Fitness
Node Fit-
ness Weight
Weighted Node
Fitness
All nodes
Without
A
0,22
0,67
0,40
k
A
=0,60
0,13
B
0,50
0,00
k
B
=1,00
0,50
C
0,50
0,50
k
C
=0,50
0,25
Fitness function must be evaluated continuously, starting from a boot process; at
boot the only etiquette behaviour that a node can complete is the one defined by E1, it
returns initially zero fitness. Evaluation of a local fitness function (that is easy to
