[ieee 2012 the 11th annual mediterranean ad hoc networking workshop (med-hoc-net) - ayia napa,...
TRANSCRIPT
Tree-Forming Schemes for Overload Control in
Wireless Sensor Networks
Charalambos Sergiou and Vasos VassiliouNetworks Research Laboratory
Department of Computer Science
University of Cyprus
Nicosia, Cyprus
Email: {sergiou,vasosv}@cs.ucy.ac.cy
Abstract—Node placement in Wireless Sensor Networks(WSNs) is frequently redundant and massive. Massive and densedeployments require efficient and effective, topology controlalgorithms in order to reduce the number of active nodes and easethe transportation of data from sources to sinks. Tree-forming
is a common strategy in topology control algorithms that canbe employed for such purpose. The most common tree-formingscheme, is shared, core-based tree, which initiates from the sink.In this work, we study how tree-forming schemes that initiatefrom sources can affect the performance of congestion/overloadcontrol algorithms. Specifically, we study source-based routingtrees in comparison with shared-, core-based trees, under trafficand resource congestion control algorithms.
Index Terms—Wireless Sensor Networks, Congestion Control,Source Based Trees, Sink Based Trees
I. INTRODUCTION
A Wireless Sensor Network (WSN) is a network com-
posed of wireless sensor nodes, capable of sensing some
phenomenon, transforming this analog data into digital and
transmitting the information to destination nodes (usually
called sinks). Due to severe power limitations their computa-
tion capability, as well as their transmission range, is limited.
A WSN comprises of a potential large set of nodes that may
be distributed over a wide geographical area indoor or outdoor.
Thus, for the transmission of data from a source (the node that
sensed the phenomenon) to a sink, the wireless sensor nodes
that lie between them, form a ”path” and data are transmitted
through it in a hop-by-hop fashion. Frequently, sensor nodes
are densely deployed near the event sources and sinks in a
redundant manner [1]. Traffic patterns in WSNs can be derived
from the physical processes that they sense. Wireless Sensor
Networks typically operate under light load and they suddenly
become active in response to a detected or monitored event.
Depending on the application this can result in the generation
of large, sudden and correlated- synchronized impulses of data
that must be delivered to a small number of sinks without
significantly disrupting the performance (i.e fidelity) of the
sensing application. This high generation of data packets is
usually uncontrolled and often leads to congestion (overflowed
buffers or packet collisions in the medium). There are a
number of protocols proposed in literature for congestion
control in Wireless Sensor Networks. A detailed analysis of a
number of them can be found in [2] and [3].
Coverage and reliability issues render massive and dense
placement of Wireless Sensor Nodes compulsory for many
applications. Massive and dense placement creates a number
of critical issues that, if not addressed correctly, are possible
to ruin completely the network’s task. Interference between
neighbor nodes, overload, and routing problems, are certain to
arise, if all nodes are allowed to transmit data each time they
sense an event. Thus, effective and efficient topology control
algorithms must be employed in order to reduce the number
of active nodes and ease the routing process [4].
Tree-forming schemes, is a common category of topology
control algorithms, that are usually employed when dense
and massive placements of sensor nodes exist. The most
common type of tree topologies, is the shared, core-based
tree, which has the sink node as the root. Networks that
employ this specific type of topology control scheme, create
a spanning tree that initiates from the sink and as a result the
initial placement and redundant connections between nodes
are severely reduced. Shared, core-based trees, also known as
sink-based trees in WSNs, are proven to be an efficient and
effective topology control solution. Sink-based trees is the term
that we will use for the rest of the paper.
In this work, we focus on congestion control algorithms and
we study whether specific tree-forming schemes can improve
the performance of congestion control algorithms when the
network faces an overload condition. In particular, we focus
on source based trees. Source-based trees are created by each
node that becomes source node (its analog part senses an
event). These trees have not been widely adopted in WSNs
since they incur excessive overhead to the network. In this
paper we study whether and how source-based trees can
improve, under specific circumstances, the performance of
congestion control algorithms. This work complements and
extents the work presented in [5].
The rest of the paper is organized as follows. In section II
we present a brief description of congestion control methods
in WSNs. In section III we describe the source and sink based
trees, while in section IV we present the evaluation and results.
Finally, we close with discussion and conclusions.
2012 The 11th Annual Mediterranean Ad Hoc Networking Workshop (Med-Hoc-Net)
978-1-4673-2039-9/12/$31.00 ©2012 IEEE 61
II. CONGESTION CONTROL METHODS IN WSNS
Congestion in WSNs happens when the offered load is
temporarily higher than the load the network resources can
process. Congestion happens either in the medium or in node
buffers. In the first case (medium) many nodes within range
of one another attempt to transmit simultaneously, resulting
in packet losses due to interference and thereby reducing
the throughput of all nodes in the area. In the other case
(buffer), node receives packets with a higher data rate than
the maximum data rate that it can transmit and the buffer of
this particular node overflows.
Currently, congestion control algorithms employ two dif-
ferent methods in order to face overload conditions: “Traffic
Control” and “Resource Control” [6]. Algorithms that employ
the“Traffic Control” method attempt to adjust the data rate of
the sources in order not to exceed the capacity of data that
nodes can store and process. This method has been employed
by the vast majority of congestion control algorithms [7], [8],
[9], [10], [11], [12], [13], [14] and it is similar to the approach
that is applied for congestion control in traditional networks.
On the other hand, algorithms that apply the “Resource
Control” method, attempt to take advantage of the dense and
massive placements of wireless sensor nodes. In this case,
when a node senses that is becoming overloaded, it informs,
through a backpressure message, the nodes that are forwarding
data through itself, to search for another forwarding node,
since it will soon be unable to accept any more packets. Thus,
algorithms that apply the resource control method, employ
nodes that are not in the initial path from the source to the
sink, in order to forward the excess data through them. The
data rate of the sources is not affected in this scheme.
III. TREE-FORMING SCHEMES
Tree-forming is an efficient and effective solution for topol-
ogy control in WSNs. The tree-forming scheme that has been
widely employed in WSNs is sink-based trees.
A. Sink-based Trees
Sink-based trees use the sink as the core node. This is
optimum in WSNs, since the sink is usually a robust node
that does not suffer from power or processing limitation issues.
Thus, sink-based trees defeat the main disadvantage of share-
core based trees which is the fact that the core node can
become a single point of failure.
In Fig. 1 the initial connectivity of the network is presented
while in Fig. 2 a sink-based tree, that resulted from the initial
topology (Fig. 1) is presented.
It is clear that creating a level-based, shared tree, leads to
a ”relaxed” topology (with a lower degree of connectivity).
This leads to fewer connections between the nodes, less
interference, and less probability of congestion occurrence.
The process for the creation of sink-based trees is analyti-
cally described in [5].
Fig. 1. Initial Network Connectivity (before topology control algorithmapplies)
Fig. 2. Sink Based Tree
B. Source-based Trees
Source-based trees introduce a completely different routing
approach in comparison with sink-based trees. The major
difference lies on the fact that source based trees, are built on
demand and only when a node becomes a source node. Thus,
each time a node becomes a source node it creates its own
routing tree from itself to sink. As stated in [5] the creation
of a properly constructed and well tuned source based trees
is not a straightforward procedure as with sink-based trees. If
the procedure is not monitored carefully it is easy to result
in a ”naive” tree, which it is not acceptable in many cases.
62
Figure 3, presents a “naive” tree.
Fig. 3. A “naive” source based tree
A “naive” tree presents several drawbacks which can be
proven fatal for network operation especially when resource
control-based algorithms apply. Resource control algorithms
base their operation, on the creation of alternative paths to
sink, in order to avoid an overloaded node. If the routing tree
is not properly constructed, the appearance of routing circles
is possible.
To construct source-based trees, which can fulfill their
purpose, we must consider a small set of “critical” parameters:
Location Awareness - Each node must be aware of its loca-
tion and the location of the sink. Localization and positioning
is a subject that attracted a lot of attention and solutions exist
in literature [15]. Location awareness is important for each
node that is going to participate in a source based tree, since
forwarding of nodes must be in the direction of the sink. Thus
each node must select as neighbor nodes, nodes that are closer
to sink than itself.
Higher Level Connection Availability - A node must be
kept on the source-based tree, only if it can communicate with
at least one node that is closer to sink than itself. In case that
higher level connection availability is not possible, then this
node cannot be part of the tree.
Number of nodes kept in neighbor table - The number of
nodes that must be kept in the neighbor table must be optimal.
In dense placements is possible to allow a large number of
nodes as neighbors. According to [16] a value of six neighbors
seems to be optimal for most topologies.
Employing these parameters in Fig. 3, the initial source
based tree transforms to the one in Fig. 4
IV. EVALUATION
To compare the performance of source-based against sink-
based trees, when resource and traffic control algorithms for
Fig. 4. A properly constructed source based tree
congestion control in WSNs are employed, we performed a
series of simulations.
A. Network Model
We assume that there is a WSN, where nodes are initially
deployed uniformly. One sink is located at a specific point
in the network. Each node is considered to be aware of its
position in relation to the sink. In our network deployment
we consider that all nodes (except the sink) are identical
and “carrier sense multiple access with collision avoidance”
(CSMA/CA) is employed as MAC protocol.
Moreover, in order to compare the performance of source-
and sink-based trees when the network faces congestion, we
employed as a resource control reaction mechanism, the alter-
native path creation scheme of Hierarchical Tree Alternative
Path (HTAP) algorithm [17].
Also, we employed as a “traffic control” mechanism, a
simple mechanism, which in case of congestion informs with a
backpressure message the sources, to reduce the data rate. By
using the “additive increase multiplicative decrease” (AIMD)
method, this algorithm attempts to adapt the data rate to the
capacity of the network. When the “traffic control” algorithm
is employed the packets are always routed through the shortest
path from each source to the sink.
B. Simulator Setup
For our simulations we employed the Prowler simulator
[18]. The radio propagation and transmission models we used
are given by:
Prec,ideal(d)← Ptransmit
1
1 + dγ(1)
where, 2 ≤ γ ≤ 4 and
63
Prec(i, j)← Prec,ideal(di,j)(1 + a(i, j))(1 + β(t)) (2)
where Ptransmit is the signal strength at the transmitter and
Prec,ideal(d) is the ideal received signal strength at distance
d, a and β are random variables with normal distributions
N(0, σa) and N(0, σβ), respectively. A node j can receive
packets from node i if Prec(i, j) > ∆ where ∆ is the
threshold.
In our simulations we used the following default simulator
parameters:
• sigmaa = 0.5• sigmaβ = 0.03• perror = 0.05• ∆ = 0.1
The rest of the parameters reflect Mica-Z node specifica-
tions, the most important of which are presented in Table I.
TABLE ISIMULATION PARAMETERS
Max Data Rate (kbps) 128Transmission Power (dbm) 2Receive Threshold (dbm) -74Transmission Current (mA) 17.4Receive Current (mA) 19.7Fragment Size (bit) 1024Buffer Size(Bytes) 512KMAC layer CSMA/CA
C. Scenarios and Results
For each of the performance metrics presented below, the
results are the average of 20 runs for 10s for each measurement
point (except for the cases that it is otherwise stated). Nodes
are uniformly placed on a square grid. All nodes communicate
only with the nodes that are one hop away. We employed
12 sources and every source is programmed to transmit at
maximum 128 packets/s, while we increase the number of
nodes on the grid.
The first metric we examine is sink throughput. Sink
throughput indicates the ability of any congestion control
algorithm, to control congestion and transmit a maximum
number of packets to the sink.
Fig. 5 presents the results when the resource control method
applies.
Results show that, when resource control is used along with
a source based tree, the sink throughput increases. This result
indicates the ability of source-based trees to provide more
alternative paths, in comparison with sink-based trees. Thus,
algorithms that employ resource control methods are favored
by the employment of source-based trees and as a result they
enhance their performance in terms of sink throughput. Sink-
based trees cannot provide as many alternative paths as source-
based trees and normally throughput is reduced. A “naive”
tree is the worst option, since routing circles and uncontrolled
number of alternative paths enhance may congestion problem
instead of relieving it.
50 100 150 200 250 300 3500
200
400
600
800
1000
1200
1400
1600
1800
Number of Nodes
Sin
k T
hro
ug
hp
ut
(pkts
/s)
Source Based Tree
Sink Based Tree
"Naive" Source Based Tree
Fig. 5. Resource Control Method: Sink Throughput (pkts/s)
Also, we notice that as the number of nodes in the network
increases, throughput also increases since more alternative
paths can be provided.
When “traffic control” method is employed, the situation is
different (Fig. 6).
50 100 150 200 250 300 350350
400
450
500
550
600
650
700
750
800
Number of Nodes
Sin
k T
hro
ug
hp
ut
(pkts
/s
Source Based Tree
Sink Based Tree
"Naive" Source Based Tree
Fig. 6. Traffic Control Method: Sink Throughput (pkts/s)
In this case, the sink throughput is significantly reduced
in comparison with the resource-control method. When a
traffic-control method applies, the network attempts to adjust
the traffic to the available network resources. Thus, since
algorithms that employ traffic control do not alter their routing
path and always use the shortest path, the sink throughput is
affected by the number of flows that cross each other. It is
notable that in this scenario the sink-based tree provides the
worst results in comparison with the other two trees. This is
normal, since when a sink-based tree is employed data flows
from each source share the same topology. On the other hand,
when source and “naive” trees are employed, each source
node creates its own routing tree from the beginning. Hence,
more nodes are employed as part of the tree and a lower
64
number of data flows cross between them. “Naive” trees, give
acceptable results concerning sink throughput, since after the
tree creation, each source finds the shortest path to the sink.
The results are worse than source-based trees, since a naive
trees create longer paths to the sink and flows cross in more
nodes in comparison.
The next parameter we study is the average delay of
successfully received packets from the sources to the sink.
Fig. 7 presents the results for the resource control method.
50 100 150 200 250 300 35020
30
40
50
60
70
80
90
100
Number of Nodes
Av
era
ge
So
urc
e t
o S
ink
De
lay
(m
S)
Source Based Tree
Sink Based Tree
"Naive" Source Based Tree
Fig. 7. Resource Control Method: Average Delay from Sources to Sink (ms)
Sink-based trees, provides less delay in comparison with
the other trees. This is normal since a sink-based tree is a
spanning tree and the position from each node to the sink is
the minimum. Between the other two trees, the average delay
when source-based tree is employed is less than that of the
“naive tree”. Source-based trees provide more routing paths in
case of congestion while “naive” trees provide longer routes
and in some cases routing circles, that increase the delay.
Fig. 8 presents the results when traffic control is employed.
50 100 150 200 250 300 35022
24
26
28
30
32
34
36
38
Number of Nodes
Av
era
ge
So
urc
e t
o S
ink
De
lay
(m
s)
Source Based Tree
Sink Based Tree
"Naive" Source Based Tree
Fig. 8. Traffic Control Method: Average Delay from Sources to Sink (ms)
In this case the overall delay is less, in all three topologies,
in comparison with the case that the resource control method
applies. This result is expected, since algorithms that employ
traffic control, always use the shortest path from source to sink
and do not re-route any packets. Between the three topologies,
sink-based trees present much better results in comparison
with the other two trees. The reason lies in the fact that sink-
based trees create a spanning tree using a shared topology. On
the other hand, source-based trees, present an increased initial
overhead, in order to create each tree from each source, while
the “naive tree”solution provides longer routes, which results
in increased delay.
Finally, we study the average energy consumption of the
network. Each energy unit is equal to 1000mW in 1 second.
Fig. 9 presents the results when resource control is used.
As it is expected, the “naive” tree presents the worst results
while the sink-based tree, the best.
50 100 150 200 250 300 3502
2.5
3
3.5
4
4.5
5
5.5
6
Number of Nodes
En
erg
y C
on
su
mp
tio
n (
un
its
)
Source Based Tree
Sink Based Tree
"Naive" Source Based Tree
Fig. 9. Resource Control Method: Average Energy Consumption (units)
This is normal, since sink-based tree creates a shared
topology and nodes follow the shortest paths in order to reach
sink. On the other hand, source-based trees create a big number
of paths and in some cases nodes follow longer routes in order
to reach sink.
Concerning the “traffic control” algorithm, the situation is
as follows.
In this case, also sink-based trees provide the best results,
but the reason is different. According to Fig. 6, when the traffic
control method applies, sink-based trees provide the sink with
the fewer packets, since the sources’ data rate is reduced.
Less packets leads to fewer transmissions and less energy
consumption. Moreover, the average delay is also smaller,
which translates into fewer hop counts.
V. DISCUSSION
Studying the results we notice that source-based trees can
provide a proper topology control solution in WSNs when
overload conditions exist. Source-based trees provide better
results in terms of sink throughput when either traffic or
65
50 100 150 200 250 300 3501
1.2
1.4
1.6
1.8
2
2.2
2.4
2.6
Number of Nodes
En
erg
y C
on
su
mp
tio
n (
un
its
)
Source Based Tree
Sink Based Tree
"Naive" Source Based Tree
Fig. 10. Traffic Control Method: Average Energy Consumption
resource congestion control algorithms apply. The drawback
is the longer delay and the higher energy consumption they
present, in comparison with sink-based trees. Therefore, we
consider that source-based trees can provide a significant
topology control solution when the application demands the
vast majority of packets to reach sink. In this case, source-
based trees can be combined with resource-control algorithms.
However, source-based trees must be carefully tuned since the
creation of a “naive” tree, can prove extremely disadvanta-
geous in the operation of the network.
Moreover, results show that when sink-based trees are
combined with resource control algorithms they can also
provide very good results, in terms of delay and energy
consumption. We believe that this solution can present a
notable overall behavior, if carefully tuned. Finally, sink-based
trees, if combined with traffic-control algorithms can provide
excellent results in terms of delay and energy consumption.
This combination has already been proposed in literature
and can be used effectively for transient congestion control
situations.
VI. CONCLUSION
In this paper, we discuss how different tree structures can
be used for overload control in WSNs. Specifically, we present
source-based trees, shared-, core-based trees (sink-based trees)
and a naive source-based tree structure and evaluate them
under resource and traffic congestion control methods. Sim-
ulation results show that not only sink-based trees, but also
source-based trees can provide efficient and effective topology
control solutions, which are, of course, application specific.
ACKNOWLEDGMENTS.
This work has been partly conducted under the European
Union Project GINSENG funded under the FP7 Program
(FP7/2007-2013) grant agreement no 224282.
REFERENCES
[1] S. Toumpis and L. Tassiulas, “Optimal Deployment of Large WirelessSensor Networks,” IEEE Transactions on Information Theory, vol. 52,no. 7, pp. 2935–2953, 2006.
[2] C. Wang, M. Daneshmand, B. Li, and K. Sohraby, “A Survey ofTransport Protocols for Wireless Sensor Networks,” IEEE Network,May/June, vol. 20, pp. 34–40, 2006.
[3] C. Wang, K. Sohraby, B. Li, and W. Tang, “Issues of TransportControl Protocols for Wireless Sensor Networks,” in Proceedings ofInternational Conference on Communications, Circuits and Systems(ICCCAS), 2005.
[4] J. N. Al-karaki and A. E. Kamal, “Routing Techniques in WirelessSensor Networks: A Survey,” IEEE Wireless Communications, vol. 11,pp. 6–28, 2004.
[5] C. Sergiou and V. Vassiliou, “Source-based Routing Trees for Effi-cient Congestion Control in Wireless Sensor Networks,” in DistributedComputing in Sensor Systems and Workshops (DCOSS), InternationalConference on (PWSN2012 Workshop), May 2012.
[6] J. Kang, Y. Zhang, and B. Nath, “TARA: Topology-Aware ResourceAdaptation to Alleviate Congestion in Sensor Networks,” IEEE Trans-actions on Parallel and Distributed Systems, vol. 18, no. 7, pp. 919–931,2007.
[7] C. Wang, K. Sohraby, and B. Li, “SenTCP: A Hop-by-Hop CongestionControl Protocol for Wireless Sensor Networks,” IEEE INFOCOM(Poster Paper), March 2005.
[8] Y. Sankarasubramaniam, O. Akan, and I. Akyildiz, “ESRT: Event-to-Sink Reliable Transport in Wireless Sensor Networks,” in MobiHoc ’03:Proceedings of the 4th ACM International Symposium on Mobile AdHoc Networking & Computing. New York, NY, USA: ACM, 2003, pp.177–188. [Online]. Available: http://dx.doi.org/10.1145/778415.778437
[9] C. T. Ee and R. Bajcsy, “Congestion Control and Fairness for Many-to-One Routing in Sensor Networks,” in SenSys ’04: Proceedings of the2nd international conference on Embedded networked sensor systems.New York, NY, USA: ACM, 2004, pp. 148–161.
[10] B. Hull, K. Jamieson, and H. Balakrishnan, “Mitigating Congestion inWireless Sensor Networks,” in SenSys ’04: Proceedings of the 2ndInternational Conference on Embedded Networked Sensor Systems.New York, NY, USA: ACM, 2004, pp. 134–147.
[11] C.-Y. Wan, S. B. Eisenman, and A. T. Campbell, “CODA:Congestion Detection and Avoidance in Sensor Networks,” inSenSys ’03: Proceedings of the 1st international Conferenceon Embedded Networked Sensor Systems. New York, NY,USA: ACM Press, 2003, pp. 266–279. [Online]. Available:http://portal.acm.org/citation.cfm?id=958523
[12] K. Sundaresan, V. Anantharaman, H.-Y. Hsieh, and R. Sivakumar,“ATP: A Reliable Transport Protocol for Ad Hoc Networks,” IEEETransactions on Mobile Computing, vol. 4, pp. 588–603, 2005.
[13] F. Stann and J. Heidemann, “RMST: Reliable Data Transportin Sensor Networks,” in Proceedings of the First InternationalWorkshop on Sensor Net Protocols and Applications. Anchorage,Alaska, USA: IEEE, April 2003, pp. 102–112. [Online]. Available:http://www.isi.edu/ johnh/PAPERS/Stann03a.html
[14] C.-Y. Wan, A. T. Campbell, and L. Krishnamurthy, “P.S.F.Q: A ReliableTransport Protocol for Wireless Sensor Networks,” in WSNA ’02:Proceedings of the 1st ACM international workshop on Wireless SensorNetworks and Applications. New York, NY, USA: ACM Press, 2002,pp. 1–11. [Online]. Available: http://dx.doi.org/10.1145/570738.570740
[15] J. Hightower and G. Borriello, “Location Systems for UbiquitousComputing,” Computer, vol. 34, pp. 57–66, August 2001. [Online].Available: http://dx.doi.org/10.1109/2.940014
[16] N. Li, J. Hou, and L. Sha, “Design and Analysis of an MST-based Topol-ogy Control Algorithm,” in INFOCOM 2003. Twenty-Second AnnualJoint Conference of the IEEE Computer and Communications. IEEESocieties, vol. 3, 2003, pp. 1702 – 1712 vol.3.
[17] C. Sergiou, V. Vassiliou, and A. Pitsillides, “Reliable Data Transmissionin Event-Based Sensor Networks During Overload Situation,” inWICON’07: Proceedings of the 3rd International Conference on WirelessInternet, Austin, Texas, October 2007, pp. 1–8.
[18] Prowler: Probabilistic Wireless Network Simulator. [Online]. Available:http://www.isis.vanderbilt.edu/Projects/nest/prowler/
66