Database Reference
In-Depth Information
same. Then, each node informed about this temporary state will add all of its
coordinate to every query that it will be asked to route and that is evaluated to
cross the node
2
in recovery failed state. The node in recovery failed state will
scan each attached coordinate in the query message looking for a coordinate
which is parent of the broken one, so that it can perform a recovery.
5 Local Learning
In this paper we improve the routing eciency by introducing a new feature
named local learning (LL). In a wireless environment uni-cast is never actually
uni-cast, in fact, whenever a node communicates with one of its neighbors the
communication is overheard by all of them, what it happens is that only the re-
cipient of the communication will listen it. In the same way, each routing request
exchanged among couples of nodes are heard by their respective neighbors. Our
idea is that overhearing neighbors may not ignore the informations they “listen
to” but they may process them instead, finding for help to give the (neighbor)
routing nodes. It may happen that a node apparently farthest from the des-
tination is aware of a node that would shorten the path, by giving back this
information to its neighbor that was routing a message through another node
it is possible that at the next routing the helping node will be chosen, and the
path will be shorten. Local is referred to the fact that what nodes learn regards
only their direct neighbors.
Every time that a packet
p
i
(data or query) with destination
c
i
is forwarded
by a sensor
d
f
(forwarder) to a sensor
d
r
(receiver), each sensor that is within the
radio range (neighbors
N
)of
d
f
is aware that the packet is being forwarded. As
a consequence each sensor in
N
can discover (i) which virtual coordinate is the
destination of
p
i
, (ii) which sensor
d
r
has been chosen towards that destination
and (iii) which is the distance of
p
i
at
d
r
. Here comes the local learning. If
any sensor in
N
,letussay
d
l
finds out that a neighbors, let us say
d
nf
with
coordinate
c
nf
would have taken
p
i
closer than
d
r
then
d
l
temporarily stores
the pair (
d
nf
,c
i
) so that when it performs the next beaconing it informs
d
f
that
a better path has been discovered. In this way, the next time that
d
f
needs to
forward a packet to a destination whose prefix is
c
i
,
d
l
will be preferred to
d
r
.
Figure 3 shows an example of local learning. Packet
p
i
with destination
011
must be routed by node
d
f
. By forwarding
p
i
to
d
l
the distance from
d
f
to the
destination is 3 while by forwarding
p
i
to
d
f
the distance is 5. Local learning
allows
n
f
to know that
d
l
is a better choice for routing packets whose destination
is
∗
011%
3
.
A possible variation of the strategy is to choose between
d
l
and
d
r
according
to a certain probability, so that possible changes in network topology and con-
sequently new possible paths can be caught even if some learning has already
∗
2
Each node can estimate if the query is likely to cross the orphan node by comparing
query destination and node.
3
% means a binary string of arbitrary length.
Search WWH ::
Custom Search