Escolar Documentos
Profissional Documentos
Cultura Documentos
Email: wangyu@ieee.org
‡ Department of Computer Science, The University of Alabama, USA
Email: yangxiao@ieee.org
Abstract—To preserve network connectivity is an important techniques (NFT) [7], which results in much high complexity
issue especially in wireless sensor network, where wireless links and cost of the algorithm. Furthermore redundant edges would
are easy to be disturbed and tiny sensors are even easy to fail be added to the topology when FLSSk locally constructs a
accidently. Therefore, it is necessary to design a fault-tolerant
network. A feasible method is to construct a k-vertex connected subgraph.
topology. In this paper, we consider k-connectivity of wireless There have been several research efforts recently on the-
network and propose a simple global algorithm (GAFTk ) which oretically studying the necessary condition for k-connected
preserves the network k-connectivity and reduces the maximal topology [8] [9] and devising approximation algorithms to
transmission power (TP). The average “ degree” expectation of the
2 construct such topologies [4]. All of above fault-tolerant
topology generated by GAFTk is O (k+3) . Based on GAFTk ,
4 topology control algorithms either use unrealistic assumptions
we propose a simple local algorithm (LAFTk ) which preserves
(i.e. the locations of all nodes and the exact distance among
k-vertex connectivity while maintaining bi-directionality of the
network. Simulation results show that GAFT/LAFT have better them are known) or involve complex calculation.
performance than other current fault-tolerant protocols. In this paper, we proposal two TC algorithms with sim-
ple calculation to preserve the network k-connectivity. The
I. I NTRODUCTION main contributions of this paper include: (a) the topologies
Fault tolerant topology control (TC) is important in wireless constructed under GAFTk and LAFTk preserve the network
networks, especially in wireless sensor network (WSN), since k-connectivity; (b) the time complexity of both GAFTk and
sensor nodes and wireless links are easy to fail and power LAFTk is low, so that nodes with limited computational and
and computational resource is limited. Many topology control power capacity can still afford these algorithms efficiently; (c)
algorithms have been designed to maintain network connectiv- the resulting topology can be converted into one with only
ity while reducing energy consumption and improving network bidirectional links.
capacity [1]. A few of them have considered the fault-tolerance The rest of the paper is organized as follows. Section II
[2], [3], like CBTCk (α) [4]. introduces our models and assumptions. Section III presents
The CBTCk (α) is an extension of CBTC from [5] to provide GAFT and its performance analysis, while Section IV gives
fault tolerant topology. In CBTC(α), every vertex increases its distributed implementation LAFT. Simulation study on
its transmission power (TP) until either the maximum angle both GAFT and LAFT is provided in Section V. Section VI
between its two consecutive neighbors is at most α (a constant concludes the paper.
depended on k) or its maximal power is reached. Therefore,
the TP would adjust to a excessive value. II. M ODELS AND A SSUMPTIONS
Fault tolerant Local Spanning Sub-graph (FLSSk ) can pre-
serve k-connectivity while maintaining bi-directionality [6]. It Assume that the nodes in network are uniformly deployed
is based on a min-max optimal centralized algorithm, Fault tol- in a c × c area. Each node has an omni-directional antenna,
erant Global Spanning Sub-graph (FGSSk ). Whether FGSSk which can adjust its TP discretely. Its transmission radius is
is k-connected that need be tested by using network flow denoted as r and its maximal value is denoted as rmax . An
undirected simple graph G = (V, E) is applied to analyze
This work is supported by the Nature Science Foundation of China under our TC algorithm, where V is the set of all nodes in the
Grant 60604029, 60702081, Nature Science Foundation of Zhejiang Province network and E is the edge set. |V | = n and | ∗ | is the size
under Grant Y106384,Y107309, Joint Funds of NSFC-Guangdong under
Grant U0735003, the SRFDP under Grant 20050335020, the R&D Program of of the set ∗. d(Nu , Nv ) denotes the received signal strength
Zhejiang Province under Grant No.2007C31038 and 111 Projects under Grant indication difference (RD) between nodes Nu and Nv . So
B07031; The work of Yu Wang was supported in part by the National Science GAFT/LAFT do not depend on any radio propagation model.
Foundation under Grant NeTS-NOSS-0721666 and by funds provided by the
University of North Carolina at Charlotte;Prof. Xiao’s work was partially A unique address ID is assigned to each node. The one-hop
1
supported by the National Science Foundation under the grant CNS-0716211. neighborhood of a node Nu , denoted as HN u
, is the set of
neighbors that Nu can directly reach by the maximal TP. We Algorithm 1: GAFT
also assume that G(V, E) is k-connected. Input : G(V, E), a simple connected graph;
Receiver can know the transmission power Pt of transmitter, Output: Gk (V, Ek ), a k-connected spanning subgraph of G;
1 begin
which is beforehand included in the hello message. And the 2 Ek := ∅; Vk := SN0 ; Vf := ∅;
receiving power Pr can be calculated from RSSI tested at the 3 while Ni ∈ Vk & Ni ∈ / Vf do
receive antenna. So RD = Pt − Pr and it can indicate the 4 Ni selects its k “closest” neighbors, i.e. SNi ;
5 if SNi ⊆ Vk then
distance d between two nodes. 6 Find a node Nj ∈ / Vk , which are closet to a node in Vf ;
7 Vk := ∪{Nj } ∪ SNi ;
III. G LOBAL ALGORITHM FOR FAULT T OLERANT–GAFT 8 Ek := ∪{e(Ni , Nl )}, Nl ∈ SNi ;
9 Vf := ∪{Ni };
The basic idea of GAFT is growing the k-connected sub- 10 Continue;
graph from a single node step by step and in each step a node
11 else
inside the subgraph invites new nodes joining the subgraph. 12 SNi = SNi ∩ Vk ;
There are two criterias for the selection of new nodes: (1) 13 Connect to k − |SNi | neighbors closest to nodes in Vk ;
adding the new nodes (or links) can improve the connectivity 14 Adjust all nodes’ TP according to algorithm 2;
of the subgraph to satisfying k-connected requirement; (2) 15 end
if the subgraph already satisfies connectivity requirement or
multiple links can improve the connectivity, select those closet
nodes. In GAFT, Vk is the set of nodes currently in the
is proved. Therefore there must be SNi * Vk . Then GAFT
subgraph, while Vf is the set of those who already connect to
runs SNi = SNi ∩ Vk and SNi ⊆ C (otherwise V1 and V2 are
at least k neighbors in the subgraph (thus no need to be con-
connected). Next Ni connects to k − |SNi | neighbors closet
sidered again). Assume GAFT begins from node N0 . Initially,
to Vk . Since these k − |SNi | neighbors respectively connect
Vk = N0 and Vf = ∅. Then N0 connects with its k closest
to nodes in Vk , there are still k links between Vk and Ni .
neighbors. If defining Ni and its closet connected neighbors
Because |C| < k, there must be at least one link between Ni
as a set SNi , we can obtain SN0 . Then we merge SN0 into
and V2 . V1 and V2 are connected.
Vk , i.e. Vk := ∪SN0 . Since N0 already has k neighbors in Vk ,
If C separates Gk into more than two disjoint components,
we then include N0 in Vf . If all node are in Vf , i.e. V = Vf ,
the above proof can be applied to every pair of components
GAFT terminates. Next, node Ni (Ni ∈ Vk ∩ Vf ) tries to
then the k-connectivity of Gk can be obtained.
connect with its k closest neighbors. Notice, Ni selects the
GAFT only needs RD. The number expectation of nodes
neighbors who can improve the connectivity of the subgraph
in the maximal cover range of a node is ne , nπ( rmax 2
c ) .
(nodes in Vk ) with higher priority. If SNi ⊆ Vk , GAFT has to
So each node only needs to communicate with ne nodes, less
find another node, Nj (Nj ∈ / Vk ), which is closest to nodes in
than n. It would reduce much calculation.
Vf . Notice this can guarantee the subgraph (Vk ) is growing.
To adjust the TP, several heuristic algorithms have been
Vk := ∪{Nj } and Vf := ∪{Ni }. If SNi * Vk , Ni only selects
proposed, such as LINT and LILT [10]. These algorithms
those neighbors in Vk (thus SNi = SNi ∩ Vk ). Then GAFT
also can be used here but we propose a quite simple one
selects the remaining k−|SNi | from those nodes who are not in
as shown in Algorithm 2 in order to to reduce the number
Vk and are closest to nodes in Vf . Assume these nodes are Nk .
of communication messages. Each node can obtain the RDs
GAFT let Vk := ∪{Nk }. Notice that in this case Ni would not
from the hello messages ofcits neighbors . According to
be included in Vf , since it still need k−|SNi | connections. This
these tested distance, each node arranges its neighbors into
whole process will terminate until all nodes in Vf . The detailed
a non-decreasing order and then calculate requisite TP level
algorithm is given as Algorithm 1. Theorem 1 guarantees the
according to Algorithm 2. An affirm message is ultimately
Gk obtained by GAFT is k-connected.
broadcast to inform its neighbors of the final TP level. Each
Theorem 1. Let G=(V, E) be a graph on V={N0 , N1 , · · · , node only broadcasts two messages, which decreases power
Nn−1 } with n≥k+1. Gk (V, Ek ) is obtained by GAFT. If G(V, consumption and interference.
E) is k-connected, then Gk is also k-connected.
Theorem 2. After running TPA, Gk is bi-directionally con-
Proof: If n = k + 1, both of G and Gk are complete nected and k-connected only if G is.
graphes. When n ≥ k + 2, suppose there is a set C ⊂ V with
In fact, TPA only adds some edges on Gk as shown in
|C| ≤ k − 1 so that Gk − C is disconnected. Let V1 and V2
Algorithm 2 and makes Gk bi-directionally connected. So TPA
be two distinct connected components of Gk − C. V1 and V2
doesn’t decrease the connectivity of Gk .
are still k-connected since G is. Suppose nodes in V1 join Vk
Before giving out Theorem 3, we show a fact. Suppose
earlier than those in V2 . Vice versa.
that node Nu has m (m ≥ k) neighbors in its maximal
Before any node in V2 joins Vk , Vk ⊆ V1 ∪ C. Since GAFT
transmission range. The probability (named as P (m, l)) that
always invites new node into Vk , there must be one moment at
Nv directly ¡connects
¢ ¡ ¢ with l neighbors out of the m ones is
which a node Ni ∈ V2 is invited into Vk . If SNi ⊆ Vk , At least
P (m, l) = ml / nl since there are totally n nodes in the
one neighbor of Ni is in V1 since |C| < k. Ni must connect
network.
to the neighbor then V1 and V2 are connected. The theorem
Algorithm 2: TP adjustment–TPA Algorithm 3: LAFT
Input : A set V containing n nodes; Input : G(V, E), a simple connected graph;
Output : G(V, E), a simple connected graph G; Output: Gk (V, Ek ), a k-connected spanning subgraph of G;
1 begin 1 begin
2 for Ni ,i ← 0 to n − 1 do 2 Ek := ∅; Each node Nu sets sets VkNu = SNu and VfNu = ∅;
3 Broadcast a Hello message with maximum TP level Ptmax . Each node Nu collects the information about HN 1 and HN 2 ;
3 u u
4 if Ni receives Hello messages then 4 Finds the k closest neighbors, i.e. SNu ;
5 Detect PrNj of these messages from Nj , j ∈ Ni1 && j 6= i 5
Nu
while Ni ∈ Vk && Ni ∈ / Vf Nu
do
and calculates d(Ni , Nj );
6 if SNi ⊆ VkNu then
6 Sort its neighbors in non decreasing list LNi ;
7 if Ni need communicates with the first k nodes in LNi then 7 Find another neighbor Nj (Nj ∈ / VkNu ) closet to VkNu ;
Nu Nu
8 Calculate: PtNi = Ptmax + Ptmax − PrNk ; 8 Vk := ∪{Nj }; Vf := ∪{Ni };
9 Broadcast PtNi by an Affirm message; 9 else
10 Receive the Affirm messages from its neighbors and know the 10 SNi := ∩VkNu ;
PtNj of its neighbors; 11 Connect with k − |SNi | neighbors closet to VkNu .
11 Set its adjusted TP level as max{PtNi , PtNj };
12 end
12 Ek := ∪{e(Ni , Nl )}, Nl ∈ SNi ;
13 Adjust all nodes’ TP according to algorithm 2;
14 end
80
2-LAFT LAFT2 when k = 2. Note that it doesn’t mean that GAFT/-
3-LAFT
70 4-LAFT FALT assume the antenna pattern is a perfect disk although
60 2-GAFT other algorithms do. In the Fig.5, the radiuses of GAFT2 are
50
3-GAFT
4-GAFT
obviously lower than those of None, CBTCk ( π3 ) and LAFT2
40 and slightly lower than those of GLSS2 and FLSS2 . The
30 radiuses of LAFT2 are obviously lower than those of None
20 and CBTCk ( π3 ) and slightly higher than those of GLSS2 and
10 FLSS2 . But it does not decrease the performance of LAFT
0 since GLSS/FLSS should examine the network connectivity
100 150 200 250 300 350 400
The number of nodes before each edge is admitted to the topology, which would
lead to great calculation and data transmission.
Fig. 3. Comparison of GAFTk and LAFTk with respect to the physical node VI. C ONCLUSION
degree (k = 2, 3, 4)
In this paper, a k-connected TC algorithm GAFT is pro-
posed to preserve k-fault tolerance. Then a distributed im-
information. But LAFT can not know information out of two plementation LAFT of GAGT is presented to adapt to some
hops so some nodes may link some farer neighbors. Therefore self-organized wireless network. These two algorithms need
the average physical degrees slightly increase as the number simple calculation and result in k-connected topology.
of nodes increases in Fig.3. Under different k, the physical We consider that the transmission range is irregular and
neighbors do not change much. GAFT/LAFT are robust under variable and do not adopt UDG model. GAFT/LAFT use RD
different connectivity. to determine the linkage relation with its neighbors. The real
distance does not effect the two algorithm. The algorithm
0.90
0.88 complexity of GAFT/LAFT is lower and they use only a few
0.86
0.84
0.82
messages.
0.80
0.78 R EFERENCES
Ratio of LD to PD
0.76
0.74 2-LAFT
0.72 2-GAFT [1] P. Santi, “Topology control in wireless ad hoc and sensor networks,”
0.70
3-LAFT ACM Comp Surveys, vol. 37, no. 2, pp. 164–194, 2005.
0.25 3-GAFT [2] F. Wang, M. T. Thai, Y. Li, X. Cheng, and D. Du, “Fault-tolerant
0.20 4-LAFT topology control for all-to-one and one-to-all communication in wireles
4-GAFT networks,” Transactions on Mobile Computing, vol. 7, pp. 322–331,
0.15
2008.
0.10 [3] M. Cardei, S. Yang, and J. Wu, “Algorithms for fault-tolerant topology in
0.05 heterogeneous wireless sensor networks,” IEEE Transactions on Parallel
100 150 200 250 300 350 400 and Distributed Systems, vol. 19, pp. 545–558, 2008.
The number of nodes [4] M. Bahramgiri, M. Hajiaghayi, and V. S. Mirrokni, “Fault-tolerant
and 3-dimensional distributed topology control algorithms in wireless
multi-hop networks,” in Proceedings of IEEE Eleventh International
Fig. 4. Comparison of GAFT and LAFT with respect to the ratio of LD to Conference on Computer Communications and Networks (ICCCN),
PD, where LD and PD refer to logical degree and physical degree 2002, pp. 392–397.
[5] L. Li, J. Y. Halpern, P. Bahl, Y. M. Wang, and R. Wattenhofer, “Analysis
TP level. In order to clearly show the results of our of a cone-based distributed topology control algorithm for wireless
multi-hop networks,” in Proceeding of ACM Symposium on Principles of
280
Distributed Computing (PODC), Newport, RI, USA, 2001, pp. 264–273.
[6] N. Li and J. C. Hou, “FLSS: a fault-tolerant topology control algorithm
for wireless networks,” in Proceedings of the 10th ACM/IEEE Annual
240
International Conference on Mobile Computing and Networking (Mobi-
Com), New York. ACM press, 2004, pp. 275–286.
200
Average radius
[7] S. Even and R. E. Tarjan, “Network flow and testing graph connectivity,”
None CBTC SIAM Journal on Computing, vol. 4, no. 4, pp. 507–518, 1975.
160 FLSS GLSS [8] C. Bettstetter, “On the minimum node degree and connectivity of a
LAFT GAFT wireless multihop network,” in The Fifth ACM International Symposium
120 on Modeling, Analysis and Simulation of Wireless and Mobile Systems
(MOBIHOC),Lausanne, Switzerland, 2002, pp. 80–91.
80 [9] M. D. Penrose, “On k-connectivity for a geometric random graph,”
Communications of the ACM, vol. 15, no. 2, pp. 145–164, 1999.
40 [10] R. Ramanathan and R. Rosales-Hain, “Topology control of multihop
100 150 200 250 300 wireless networks using transmit power adjustment,” in Proceedings of
The number of nodes the IEEE Conference on Computer Communications, Tel Aviv, Israel,
vol. 2, 2000, pp. 404–413.
Fig. 5. Comparison of None, CBTCk ( π3 ), FGSS2 , FLSS2 , GAFT2 and
LAFT2 with respect to the average radius (k = 2)