Sie sind auf Seite 1von 10

Interference Effects on the OLSR Protocol:

NS-2 Simulation Results


Dang Quan Nguyen Pascale Minet

INRIA Rocquencourt, Domaine de Voluceau INRIA Rocquencourt, Domaine de Voluceau


Rocquencourt - B.P. 105 Rocquencourt - B.P. 105
78153 Le Chesnay Cedex, France 78153 Le Chesnay Cedex, France
dang-quan.nguyen@inria.fr pascale.minet@inria.fr

Abstract— Through their intrinsic qualities, mobile ad- Quality of service routing protocols usually use the
hoc networks enable a rapid deployment, without needing bandwidth reservation mechanism to provide an end-to-
a pre-built infrastructure, and are self-adaptive. Ad-hoc end quality of service (QoS) [3], [4]. We will show in this
nodes are wireless and they transmit at the same frequency paper the impact of the interference phenomenon on the
within the same radio network. The transmission of a
bandwidth effectively granted by QoS routing protocols
node can interfere with the transmission of other nodes if
at the receivers, the carrier to interference ratio is lower
and on the bandwidth amount that should be reserved.
than a threshold value. This article deals with the radio A flow can consume more bandwidth than the effective
interferences in mobile ad-hoc networks. We first show, bandwidth amount requested, due to the interference it
by using simulations under NS2, that interferences can generates along its path. Therefore, the bandwidth of
disrupt the bandwidth granted to already established flows. already accepted flows may significantly decrease as a
We propose a bandwidth reservation model which ensures new flow is introduced. Hence the interference problem
that the interferences produced by a new flow do not affect must be accounted for by QoS routing protocols.
the bandwidth granted to the already accepted flows. We In this paper, we propose an interference-aware band-
also show that link hysteresis can help a proactive routing
width reservation model that operates at the routing level.
protocol like OLSR to reduce the interference effect, thus
providing more stable routes. While other bandwidth reservation protocols like IN-
SIGNIA [4] or BRuIT [3] use reactive routing protocols
I. I NTRODUCTION (e.g. DSR, AODV) to find a route, our protocol takes
Multi-hop wireless ad-hoc networks can be quickly advantage of the proactive routing protocol OLSR [5] to
deployed and nodes adapt themselves to network topol- find a route more rapidly.
ogy changes. Such a network does not need base stations We also address the effect of link hysteresis on com-
which are required in other types of wireless networks munication loss rate. In order to calculate the OLSR
(e.g.: cellular networks,...). Each node of an ad-hoc routing table, each node advertises its links with a
network can move freely and keep its communication subset of its neighbors to all nodes in the network.
as long as network connectivity is kept. Nodes use the It turns out that many among those links are unstable
same radio frequency, with limited transmission ranges, due to interferences, lack of correct reception of control
over the whole ad-hoc network. Packets are relayed to messages,. . . We will show that link hysteresis helps the
reach their multi-hop destination. If we assume that OLSR routing protocol to provide more stable routes.
the medium access problem is solved by an efficient This paper is organized as follows. In section II, we
MAC protocol (e.g. CSMA/CA) and environment spe- define our model for computing the bandwidth amount
cific problems (ex.: obstacles, multi-path fading) are that must be reserved in the presence of interferences. We
neglected, there is still a problem to solve in such give a summary of the OLSR routing protocol and de-
networks. How does a routing protocol account for scribe our bandwidth reservation protocol in section III.
the interferences: communication between two adjacent Results of the NS-2 simulations with the OLSR protocol
nodes can be altered by distant nodes transmission. Such and our reservation model are discussed in section IV.
a protocol is said interference aware. Section V presents the effect of link hysteresis on loss

470
rate. Finally, we conclude in section VI. belong to the interference area of at most 5 nodes such
as in a straight line configuration.
II. I NTERFERENCE AWARE BANDWIDTH Another interference model for bandwidth reservation
RESERVATION MODEL in ad-hoc networks has been proposed in the protocol
Researches have been carried out in order to define BRuIT [3]. In that model, the interference area of a
an interference model for ad-hoc networks. [1], [2] source node s, where nodes are affected by interferences
deal with the network capacity at the physical layer, generated by s, includes only nodes at a distance less
in the presence of radio interference. In this paper, than or equal to l, measured in number of hops, from s.
we propose an interference aware bandwidth reservation Therefore, for a flow that requires k bps, BRuIT makes
model that operates at the routing level. This simple a bandwidth reservation with the quantity of bandwidth
model computes the bandwidth consumed by flows, con- k bps not only along the path of that flow but also at
sidering the interference influence. This model provides all nodes within l hops from the path. The model does
the bandwidth reservation with an overestimation of the not take into account the path length parameter and can
bandwidth required by flows. We assume that the radio not prevent a new bandwidth reservation from interfering
transmission spreading is unbounded: radio interferences with already accepted reservations.
produced by one single transmission disturb all nodes in Our bandwidth reservation model, B = k ∗ dist, takes
the ad-hoc network. Thus, each node is considered as an into consideration the path length parameter, dist. This
interference generator by all other nodes. is to ensure that all transmitting nodes are counted.
When a source node S wants to transmit a flow to a The longer the path is, the more transmitting nodes
destination node D at distance dist from S , measured along that path. Thus, the interference level increases
in number of hops, and requires a bit rate of k bps, the proportionally.
bandwidth amount to be reserved B is:
III. BANDWIDTH RESERVATION WITH OLSR
B = k ∗ dist (1) In this section, we introduce the bandwidth reservation
mechanism in the OLSR protocol, according to our
This quantity of bandwidth, B , is reserved at each node
model described in the section II. We have done simula-
of the network by using a signaling protocol.
tions under the NS-2 simulator to show the interference
In this model, the bandwidth amount for the reser- effect on the bit rate of flows at the receiver. Simulations
vation is equal to the product of the requested amount also show that bandwidth reservation is compromised if
of bandwidth k by the path length dist. This can be the interference aspect is not accounted for.
explained because any node in the path is considered as
an interference generator with equal interference level. A. OLSR protocol
However, the distance parameter dist is not totally OLSR (Optimized Link State Routing) is an optimiza-
appropriate to provide the reservation mechanism with a tion of a pure link state routing protocol. It is based
good estimation of the interferences generated by flows. on the concept of multipoint relays (MPRs) [5]. First,
There are scenarios where a flow f1 takes a path shorter using multipoint relays reduces the size of the control
than a flow f2 , but there are more neighbour nodes messages: rather than declaring all its links to all nodes
along the path of the former than of the latter. Thus, in the network, a node declares only the set of links with
the interferences caused by flow f1 affect more nodes its neighbors that are its “multipoint relays”. The use of
than these of flow f2 . A more realistic approach to MPRs also minimizes flooding of control traffic. Indeed
compute the bandwidth to be reserved should take into only multipoint relays forward control messages. This
consideration the number of interfered nodes along the technique significantly reduces the number of retransmis-
path. sions of broadcast control messages [6]. The two main
With some additional assumptions, this model can be OLSR functionalities, Neighbor Discovery and Topology
adapted as follows: Dissemination, are now detailed.
1) Neighbor Discovery:
(
k ∗ dist if dist ≤ 5
B= Each node must detect the neighbor nodes with which it
5∗k if dist > 5
has a direct link. For this, each node periodically broad-
The interference radius is assumed to be equal to 2 times casts Hello messages, containing the list of neighbors
the coverage radius. Moreover, any node is supposed to known to the node and their link status. The link status

471
can be either symmetric (if communication is possible in B. OLSR protocol enhanced with bandwidth reservation
both directions), asymmetric (if communication is only We now propose to enhance the OLSR protocol in
possible in one direction), multipoint relay (if the link order to support QoS routing. This QoS routing is based
is symmetric and the sender of the Hello message has on bandwidth reservation. The OLSR protocol enhanced
selected this node as a multipoint relay), or lost (if the with bandwidth reservation includes three phases: (1)
link has been lost). The Hello messages are received by resource reservation, (2) data/voice/video transfer and (3)
all one-hop neighbors, but are not forwarded. They are resource release. Resource reservation uses the OLSR
broadcast once per refreshing period “Hello interval”. routing protocol to find the route to the destination,
Thus, Hello messages enable each node to discover its bandwidth is reserved on all nodes of the network
one-hop neighbors, as well as its two-hop neighbors. according to the model described in section II. Once the
This neighborhood and two-hop neighborhood informa- bandwidth is reserved, QoS routing ensures that during
tion has an associated holding time, after which it is no the transfer phase all packets belonging to that flow
longer valid. On the basis of this information, each node follow the same route to the destination. When there
independently selects its own set of multipoint relays is no more packet to transmit, the resources reserved for
among its one-hop neighbors in such a way that the this flow are released.
multipoint relays cover (in terms of radio range) all two-
1) Resource reservation:
hop neighbors (see [5] for an algorithm example). The
In this section, we assume that the available bandwidth
multipoint relay set is computed whenever a change in
is the same for all nodes in the network. Hence if the
the one-hop or two-hop neighborhood is detected. In
bandwidth available at the source allows the acceptance
addition, each node M maintains its “MPR selector set”.
of the new flow, it is the same for all nodes in the
This set contains the nodes which have selected M as
network. We will see in section III.D how to remove
a multipoint relay. Node M only forwards broadcast
this assumption.
messages received for the first time from one of its MPR
Nodes can approximately determine their local avail-
selectors.
able bandwidth in different ways. This feature is inves-
To avoid the use of bad quality links (e.g. unstable tigated in section III.C.
links), OLSR monitors the link quality according to the In the resource reservation phase,
algorithm described in section V. • an amount of bandwith is reserved at all nodes in
the network for the flow f considered.
2) Topology Dissemination: • the route R from the source to the destination is
Each node of the network maintains topological infor- determined and a soft-state entry associated with f
mation about the network obtained by means of TC is created at each node on R.
(Topology control) messages. Each node M selected as a
When a node S wants to start a flow to a node D
multipoint relay, broadcasts a TC message at least every
at a rate of k bps, the reservation mechanism locally
“TC interval”. The TC message originated from node M
computes the bandwith amount to be reserved, B =
declares the MPR selectors of M . If a change occurs in
k ∗ dist, where dist is the distance, in number of
the MPR selector set, the next TC can be sent earlier.
hops, between S and D , given by the OLSR routing
The TC messages are flooded to all nodes in the network
table. If the residual bandwidth at S is greater than the
and take advantage of MPRs to reduce the number of
bandwidth amount B , then the flow is accepted. S sends
retransmissions. Thus, a node is reachable either directly
a bandwidth reservation (BW RESV) message containing
or via its MPRs. This topological information collected
the value of the bandwidth quantity B to inform all
in each node has also an associated holding time.
nodes of the reservation. This message is broadcast to
The neighbor information and the topology informa- the entire ad-hoc network by the OLSR routing protocol
tion are refreshed periodically, and they enable each in the same way as a topology control (TC) message.
node to compute the routes to all known destinations. Every node in the network updates its residual bandwidth
These routes are computed with Dijkstra’s shortest path amount by decreasing it by B , because according to our
algorithm [7]. Hence, they are optimal as concerns the model the interference is supposed to affect all nodes.
number of hops. The routing table is computed whenever For each accepted flow f , the reservation mechanism
there is a change in neighborhood or topology informa- also determines thanks to OLSR the route R from the
tion. source to the destination. This route R is used for the

472
transfer phase. A soft-state entry associated with f is f ’s transmission, or
created or updated at each node on R when the node - when the soft state of f at the destination node
receives a marked packet. The source node marks the expires.
first packet of the flow f when it starts to transmit f In the first case, the source node releases the band-
after bandwidth reservation or after any modification of width amount B that f has reserved at all nodes in
the route R. At node Ni , the soft-state entry associated the network. It broadcasts a bandwidth free (BW FREE)
with f contains the source S , the next hop Ni+1 , the message to all nodes in the ad-hoc network. This mes-
previous hop Ni−1 , the destination D and the distance sage contains the bandwidth amount B and the flow
between Ni and D . Only nodes on the route R have a identifier fid of f . BW FREE is broadcast in the same
soft-state for the flow f , all other nodes do not. They way as a BW RESV or a TC message. Upon reception
only maintain the residual bandwidth amount. of BW FREE, a node updates its residual bandwidth by
In scenarios where topology changes cause the path adding B to it. Nodes on the route R remove their soft-
length to change, the routing protocol must adapt its state entry for fid.
bandwidth reservation mechanism in function of the In the second case, the destination node of flow f
path length variation. This topic is discussed in sec- is in charge of releasing the bandwidth granted to f . It
tion III-B.3. proceeds as the source in the first case.
2) Data/Voice/Video transfer: Let us consider the behavior of a node Ni on the
Once the flow is accepted into the network, all packets route R detecting that the link with its next hop N i+1 is
of that flow follow the route R that has been determined broken. We distinguish two cases:
by the resource reservation mechanism. Therefore, the • either the node Ni knows how to reach the destina-
route length remains unchanged and the flow consumes tion with the same distance as R, Ni locally repairs
the amount of bandwidth that was initially reserved. the route R by choosing a new next hop Ni+1 0 and
This phase can be realized using source routing or 0
marking the first packet of f it sends to Ni+1 . Thus
per-hop routing. With source routing, once the route R node Ni+10 and all the following nodes that do not
is found, R is included in all the outgoing packets at belong to R create a soft state for f . This local
the source. Intermediate nodes use this information to repair is transparent to the source.
forward packets. Including R in every packet generates • or the node Ni informs all source nodes that
overheads. Therefore the flow actually consumes more have flows going through this link of the inter-
bandwidth than the requested bandwidth amount k . ruption, by sending them a unicast route down
We adopt the per-hop routing which is the following. (ROUTE DOWN) message. This message is for-
All packets of f are forwarded by node Ni to node warded backward to the source. Upon receipt of
Ni+1 . Ni+1 is in the soft-state of node Ni , it has this message a node removes the broken link from
been determined in the reservation phase. As long as its topology table. A source can choose to relocate
all links (Ni , Ni+1 ) are maintained, all packets of f the flow on a new route by reservation, or to end
follow the same route R. We deal with broken links in the flow as described above.
section III-B.3.
If the source chooses to relocate the flow, then it has
The associated soft-state entry at a node N is refreshed
to modify the reserved bandwidth amount according to
everytime N receives a packet of f . If there is no
the length of the new route. Let ∆d be the difference of
refresh during the time soft-state validity time (SSVTime),
the lengths, expressed in number of hops:
then the soft-state entry expires. The node removes the
expired soft-state and does not recover the bandwidth
∆d = distnew route − distold route (2)
amount that has been reserved for f . In other word, soft-
state can be seen as an indication whether the node N If ∆d 6= 0, the source node broadcasts a bandwidth
belongs to the route R of f . update (BW UPD) message to all nodes in the network.
3) Resource release and reservation renewal: BW UPD is broadcast in the same manner as a TC
Resource release mechanism makes sure that nodes in message. It contains the quantity of bandwidth ∆B to
the network recover the bandwidth that has been reserved be added or removed at each node, where ∆B = k ∗∆d.
for flow f but is no longer used. This mechanism is In all cases, whatever the value of ∆d, the source node
invoked in two ways: marks the first packet of f transmitted on the new route
- when the source node explicitly notifies the end of to create/update soft-state entries.

473
We now investigate the mechanisms enabling nodes to The simulation results closely match with the analysis
obtain the amount of maximum available bandwidth. results of Bianchi’s model.
3) Dynamic estimation of local available bandwidth:
C. Maximum available bandwidth and residual band- A non-intrusive bandwidth estimation method has been
width introduced by Zhang and al. [9]. The model is based on
the observed signal-to-noise ratio (SNR) to estimate the
There exist a number of methods for bandwidth esti- residual bandwidth at MAC layer.
mation in the literature [10], [11], [12]. However, they This method can be used to dynamically and locally
are created for the wired networks and can hardly be monitor the local available bandwidth (LAB). The band-
applied to the ad-hoc wireless networks. Indeed, the width available at one node can differ from bandwidth
maximum bandwidth is much less in wireless environ- available at another node. We will show in section III-D
ment and ad-hoc nodes have limited battery capacity. It how our reservation algorithm can account for these
is then preferable not to introduce overhead to estimate different values.
available bandwidth amount like in the wired networks.
Node mobility makes it even more difficult to estimate D. Computation of the available bandwidth
the local bandwidth. Below we enumerate some simple With the dynamic estimation of local available band-
methods to estimate the amount of available bandwidth width method, the LAB can differ from one node to
in the ad-hoc wireless networks. another. Hence, we define the available bandwidth in the
1) Manual configuration: network as the minimal value among the LABs.
This is perhaps the simplest way to estimate the max- To make a reservation based on the minimum available
imum available bandwidth in an ad-hoc network. This bandwidth, a node must acquire the available bandwidth
method assumes that the bandwidth available at a node of all other nodes in the network. The following algo-
is the same for all nodes in the ad-hoc network. This rithm is used to perform this task.
available bandwidth is measured between two neigbor We assume that all nodes compute their LAB using
nodes. One can set up a two-nodes scenario and de- one of the three methods above. Nodes add an extra
termine the maximum bandwidth Bmax on the link by information which is their LAB value to each Hello
increasing the transmission rate of the source node until message. Each MPR broadcasts in its TC the LAB value
stabilization of the reception rate at the destination node. of each MPR selector. Therefore, the LAB value of each
All other nodes are kept silent during this operation. node is periodically advertised to all other nodes. With
Then the maximum available bandwidth is configured to our reservation model, each node computes the available
Bmax on all nodes participating in this ad-hoc network. bandwidth as the minimum of those LAB values, and uses
The residual bandwidth is the difference between B max this value in the reservation protocol.
and the sum of all bandwidth amounts that have been
reserved. IV. NS-2 SIMULATIONS

This method can be applied to scenarios where no new A. Simulation parameters


node shows up during an existing session. Otherwise, We have performed simulations under NS-2. The
such node should acquire the knowledge of Bmax and OLSR routing protocol and the IEEE 802.11b MAC
the bandwidth amounts that have been reserved, before protocol are used. At the physical layer, the transmission
making any resource reservation. power Pt is constant. Nodes have a transmission range
2) Bianchi’s 802.11 DCF analysis and parameters: of 250m. Within this range, they are able to correctly
Bianchi [8] has analyzed the 802.11 Distributed Coordi- receive packets from each other. This parameter in NS-
nation Function (DCF) used in ad-hoc mode. This anal- 2 is derived from the RXThresh (Receive Threshold)
ysis provides an estimation of the maximum bandwidth parameter. Between 250m and 500m, nodes are capable
in different configurations of an ad-hoc network: number to detect packets but unable to correctly receive them.
of nodes, basic access vs. RTS/CTS access. This can be Thus, the wireless channel is sensed busy (signal is con-
applied to any ad-hoc network with the appropriate pa- sidered as Interference) should there be any transmission
rameters to estimate the value of the maximum available inside this range. This value (500m) is calculated from
bandwidth. the CSThresh (Carrier Sense Threshold) parameter. The
We have done simulations under NS-2 to determine reception power (RxPr) depends on the distance between
the maximum available bandwidth at the MAC layer. the sender and the receiver and on other parameters,

474
according to the radio propagation model being used
(we use the Two-Ray-Ground propagation model). If
the RxPr/Interference ratio is lower than a Capture
Threshold (CPThres ), then a collision occurs. We use
the default values of these parameters given in table I.
The RTS/CTS option is activated in the MAC layer.
Each simulation lasts for 600s. Flows are CBR (constant-
bit-rate). The size of all packets is set at 1 KB. In
order to simplify the simulation of the model, we use
the manual configuration method to obtain and impose Fig. 1
the maximum available bandwidth to all nodes. The 8 NODES , 3 FLOWS IN Fig. 2
maximum available bandwidth between two neighbour PARALLEL 9 NODES , 2 CROSSED FLOWS

nodes using OLSR is 1.3 Mbps, as tested under NS-2.


We have used a simple scenario with only two nodes (one
sender transmits a CBR/UDP flow to one receiver) that
300 Kbps. At time t = 100s, node F starts to transmit
are within the transmission range of each other. The MAC
to node I a flow, f1 , at 300 Kbps. Node A stops its
bandwidth (bandwidth ) parameter is 2 Mbps. But the
transmission at time t = 500s, node F at t = 550s.
maximum bit rate recorded at the application layer of the
receiver node is only 1.3 Mbps. This value is therefore 1) Results without bandwidth reservation:
known to all nodes participating in this ad-hoc network Flows start and stop their transmission without regard to
configuration as the maximum available bandwidth. the available bandwidth of nodes. In the first scenario,
there is not sufficient bandwidth left for the admission
TABLE I of f2 at any node of the network. Thus, when f2 starts
PARAMETERS OF WIRELESS PHYSICAL LAYER . its transmission, interferences cause the bandwidth of
reserved flows, f0 and f1 , to decrease (cf. Figure 3). This
Parameter Value significantly affects already accepted flows: a bandwidth
Pt (W) 0.28183815 decrease of 30% for f0 , and more than 15% for f1 .
RXThresh (W) 3.652e-10
The transmission bit rate of f2 only reaches 200 Kbps
CSThresh (W) 1.559e-11
CPThresh (db) 10.0 whereas 500 Kbps are required. The flows f0 and f1
MAC bandwidth (Mbps) 2.0 meet again their bandwidth requirements at time t =
350s when f2 stops its transmission.
B. Static scenarios 1200
f0
f1
f2
Two scenarios have been considered. OLSR is used 1000
total

without link hysteresis. The first scenario includes


8 nodes (cf. Figure 1) on a surface of 1500x1500 m2 . 800

Three CBR flows are transmitted, using three parallel


bandwidth (Kbps)

paths. Flows are disjoint: no node is visited by more than 600

oneAtflow.
timeThis
t =scenario hasAbeen
50s, node used
wants to by BRuITa [3]
transmit in
flow,
its
f0 , simulations.
to node C, at 250 Kbps. Node G wants to transmit, 400

at time t = 100s, a flow, f1 , to node H at 600 Kbps.


200
Node D starts to transmit a flow, f2 , to node F, at time
t = 150s, with a bit rate of 500 Kbps. Node A stops 0
its transmission at t = 400s, G at t = 500s, and D at 0 100 200 300
time (s)
400 500 600

t = 350s.
The second scenario includes 9 nodes on a Fig. 3
1500x1500 m2 surface (cf. Figure 2). The two CBR B IT RATES WITHOUT RESERVATION IN SCENARIO 1
flows have one single common node (node C) which
relays packets on behalf of both flows. At time t = 50s, In the second scenario, the situation is similar to
node A starts to transmit a flow, f0 , to node E, at the first one. But the residual bandwidth amount is

475
TABLE II
very limited as more interference sources exist in this
F LOW PARAMETERS FOR SCENARIOS 1 AND 2.
scenario: each flow is transmitted at 300 Kbps on a 4-
hop path. When both flows are active at the same time, Scenario 1 Scenario 2
their effective bandwidth is reduced about 50%: the bit Reservation model B = k ∗ dist
rate obtained for each flow is 150 Kbps. f1 reaches its Total bandwidth (Kbps) 1300
Flow f0
required bit rate of 300 Kbps after f0 has finished its
k (Kbps) 250 300
transmission (cf. Figure 4). dist (hops) 2 4
B (Kbps) 500 1200
600 Residual bandwidth (Kbps) 800 100
f0
f1
total
Flow f1
500
k (Kbps) 600 300
dist (hops) 1 4
B (Kbps) 600 1200
400
Residual bandwidth (Kbps) 200 not accepted
bandwidth (Kbps)

Flow f2
300
k (Kbps) 500 –
dist (hops) 2 –
200
B (Kbps) 1000 –
Bandwidth left after reser- not accepted –
100 vation (Kbps)

0
0 100 200 300 400 500 600
time (s)

Fig. 4 the bandwidth requirement of f2 . The flow f2 cannot


B IT RATES WITHOUT RESERVATION IN SCENARIO be accepted. f0 and f1 have their transmission bit rates
2 left unchanged during their session.

1200
2) Results with bandwidth reservation: f0
f1
f2
The goal of the bandwidth reservation is to provide the total
1000
requested quality of service to the accepted flows, despite
the acceptation of new flows. In our study, the quality of 800

service parameter is the bandwidth. The network should


bandwidth (Kbps)

be able to provide this guarantee as long as the accepted 600

flow remains active. A new bandwidth reservation must


be accepted only if the residual bandwidth on the path 400

is still sufficient to satisfy the requirement, and radio


interferences produced by that flow do not affect the 200

already accepted bandwidth reservations. Simulations


have been performed with the previous two scenarios, 0
0 100 200 300 400 500 600
time (s)
and the OLSR protocol enhanced with a bandwidth
reservation mechanism (see table II). Fig. 5
Results obtained with the first scenario are shown in B IT RATES WITH RESERVATION IN SCENARIO 1
Figure 5. At time t = 50s, f0 starts successfully, and
transmits at 250 Kbps. The path length of f0 is 2 hops.
Therefore, the OLSR protocol makes a reservation for
Results obtained with the second scenario are shown
500 Kbps at every node in the network. The bandwidth
in Figure 6. They show that f1 is not accepted into
remaining after this reservation is 800 Kbps, as the
the network, because bandwidth is not sufficient and f 0
maximum bandwidth capacity is 1300 Kbps. The
continues to transmit at 300 Kbps until the end of its
flow f1 can also be accepted with its transmission at
session.
600 Kbps on a 1-hop path. The bandwidth left after this
reservation is 200 Kbps which is not enough to meet

476
600 TABLE III
f0
f1
total T RANSMITTED FLOWS
500 fid src dst k dist initial route
1 5 56 50 3 5, 50, 18, 56
400 2 24 12 60 4 24, 69, 61, 27, 12
3 68 14 70 2 68, 65, 14
bandwidth (Kbps)

4 4 33 80 3 4, 2, 19, 33
300
5 36 59 90 3 36, 68, 0, 59

200

100 rate at the destination nodes. Results show that the


average loss rate without reservation is 6.45% whereas
0
0 100 200 300 400 500 600 the average loss rate with reservation is only 1.12%.
time (s)

Fig. 6
120
flow 1
B IT RATES WITH RESERVATION IN SCENARIO 2 flow 2
flow 3
flow 4
100 flow 5

80
C. Generic scenarios
In this section, OLSR is used with link hysteresis as bandwidth (kbps)
60

recommended in the RFC. We have simulated with NS-2


40
a generic configuration. This configuration includes 70
nodes randomly located in a 1000x1000 m2 area. Five
20
CBR flows start and stop at different times. We compare
the bandwidth obtained by each flow without and with 0
bandwidth reservation. Simulation results are depicted 0 100 200 300
time (s)
400 500 600

in figures 8 and 9 respectively.


Fig. 8
S IMULATION RESULTS WITHOUT BANDWIDTH
Network Configuration RESERVATION
1000
17 2 19 node
flow 1
900 44 26 8 flow 2
4 flow 3
flow 4
53 42
800 38 39 36 34 flow 5

23
16
3352
35 0
15 As further work, we will study the behavior of the
68
700
37
3
58
5
protocol in presence of node mobility.
65
600 64
32
9 30 59
y (m)

500 45 24
10

400
63
5057
V. L INK HYSTERESIS AND LOSS RATE
14
13
300 2225 28 49 To avoid packet loss, the OLSR RFC [5] recommends
51 69 54 47
46 55
200 31
20
1
12 18 not to use links of bad quality and links oscillating very
41 56 66 48

100 61
43
frequently between the up and down states. Link hystere-
27 6
7 21 62
29
11 67
40
60
sis has been introduced to enhance neighbor discovery
0
0 100 200 300 400 500
x (m)
600 700 800 900 1000
by detecting and avoiding the use of such links. In this
section, we focus on link hysteresis. We first show how
Fig. 7 link hysteresis works. We then evaluate by simulation
N ETWORK CONFIGURATION WITH 70 NODES the benefit brought by link hysteresis on the loss rate.
A. Link hysteresis principles
With bandwidth reservation, we observe that the As described in [5], link hysteresis monitors for any
bandwidth obtained by each flow is smoother than link its quality, denoted L Quality . Only links of good
without bandwidth reservation. We measure the loss quality can be used by data/voice/video transfer. As a

477
100
A packet is detected lost after a gap in the packet
flow 1

90
flow 2
flow 3
sequence number or after a long period of silence.
flow 4
flow 5
80

70
The default values of link hysteresis parameters are:
TABLE IV
bandwidth (kbps)

60

50 L INK HYSTERESIS PARAMETERS


40 Parameter Value
30
Hyst Low 0.3
Hyst High 0.8
20 Hyst Scaling 0.5
10

0
0 100 200 300 400 500 600 B. Simulation results
time (s)
We consider the generic static scenario used in
Fig. 9 subsection IV-C. We assume that the OLSR protocol
S IMULATION RESULTS WITH BANDWIDTH is enhanced with resource reservation as defined in
RESERVATION section III. We compare the loss rate measured at the
destination obtained with and without link hysteresis.
In the simulations, we use the default values of link
hysteresis parameters.
consequence a link enters a pending state before being
established:
The results depicted on figure 10 are obtained without
• A link is established if its quality is strictly higher
link hysteresis. They have to be compared with those
than Hyst High. of figure 9. Moreover, the average loss rate is equal
• An established link is dropped (i.e. becomes pend-
to 4.97%, compared with 1.12% obtained with link
ing) if its quality is strictly smaller than Hyst Low, hysteresis. Hence link hysteresis is a powerful tool to
with cope with interferences.
Hyst Low ≤ Hyst High
100
flow 1
• An established link such that 90
flow 2
flow 3
flow 4
flow 5
Hyst Low ≤ L Quality ≤ Hyst High 80

70
remains in the same state.
bandwidth (kbps)

60

Notice that the condition to establish a link is stronger 50

than the condition to drop it. A link is dropped as soon 40

as L Quality decreases under Hyst Low or a timer 30

expires. 20

If the link layer does not provide information on


10
the quality of links, for any link L, L Quality can be
0
computed by an exponentially smoothed sliding average 0 100 200 300
time (s)
400 500 600

as follows:
• on the first time an OLSR packet is received on L, Fig. 10
L Quality is set to Hyst Scaling ; S IMULATION RESULTS WITH BANDWIDTH
• on receipt of an OLSR packet on L: RESERVATION AND WITHOUT LINK HYSTERESIS

L Quality = (1 − Hyst Scaling) ∗ L Quality


+Hyst Scaling Moreover, simulation results show an interesting
property provided by link hysteresis. The routes
• on detection of a lost OLSR packet on L:
taken by flows tend to move apart in order to
L Quality = (1 − Hyst Scaling) ∗ L Quality avoid interferences. We have simulated a 50-nodes

478
configuration with 3 flows (see figure 11). Initially the Simulation results show that with bandwidth reserva-
3 routes are very closed. Interferences between flows tion, the bandwidth effectively granted to flows is not
contribute to decrease the link quality. Link hysteresis subject to high variations unlike without reservation.
and the OLSR routing tend to replace bad quality links These results are made possible by means of link hys-
by better ones. The routes are updated accordingly. teresis. By detecting bad quality links, link hysteresis
avoids the use of such links.
Moreover link hysteresis is very useful to cope with
Network Configuration
1000 10 node
the interferences. It allows the routing protocol to update
27 flow 1 initial
900
28 41
17
35
43
47
flow 2 initial
flow 3 initial
the routes to minimize the interference effects between
45 24
800
18
26
0
7 25
48
flow 2 1st re-route
flow 2 2nd re-route flows.
3
700 6
11 1
Routing enhanced with bandwidth reservation allows
600
2
15
4
to reduce the loss rate as shown by our simulations.
46 5
30
y (m)

500
12
19 16
R EFERENCES
21
400 29
31
49
[1] K. Bertet, C. Chaudet, I. Guérin-Lassous, L. Viennot: Impact of
300 38
36 8 40
Interferences on Bandwidth Reservation for Ad-hoc Networks: a
44
200 20 First Theoretical Study, IEEE Symposium on Ad Hoc Wireless
39 33
100 14 37
Networks (SAWN 2001), San Antonio, USA, November 2001.
4234
22 32
13
[2] R. Hekmat, P. Van Mieghem: Interference in Wireless Multi-hop
0 23 9
0 100 200 300 400 500 600 700 800 900 1000 Ad-hoc Networks and its Effects on Network Capacity, accepted
x (m) paper for Wireless Network Journal, special issue on Ad-hoc
Networking, February 2003.
Fig. 11 [3] C. Chaudet, I. Guérin-Lassous: BRuIT: Bandwidth Reservation
N ETWORK CONFIGURATION WITH 50 NODES under InTerferences influence, in European Wireless (EW)
pp.466-472, Firenze, Italy, February 2002.
[4] S-B. Lee, G-S. Ahn, X. Zhang, A.T. Campbell: INSIGNIA: An
IP-Based Quality of Service Framework for Mobile ad Hoc
Let us consider flow 2 whose source is node 15 and Networks, in Journal of Parallel and Distributed Computing,
destination is 20. As we observe in the simulation, the NN ◦ 60, pp.374-406, 2000.
initial route taken by flow 2 is the shortest one according [5] C. Adjih, T. Clausen, P. Jacquet, A. Laouiti, P. Minet,
P. Muhlethaler, A. Qayyum, L. Viennot: Optimized Link State
to the OLSR routing protocol. It consists of nodes 46 and Routing Protocol, RFC 3626, IETF, October 2003.
19. Because of the interferences on node 19, the route [6] A. Qayyum, L. Viennot, A. Laouiti: Multipoint relaying: An
is updated to avoid that node. This second route goes efficient technique for flooding in mobile wireless networks,
through nodes 29, 49, 36. In the same way, node 36 is Research Report N ◦ 3898, INRIA Rocquencourt, France, March
2000.
later avoided because of the interferences. The final route [7] E.W. Dijkstra: A Note on Two Problems in Connection with
goes through nodes 29, 44 and 39. Graphs, Numerische Mathematik, 1, pp.269-271, 1959.
[8] G. Bianchi: Performance Analysis of the IEEE 802.11 Dis-
VI. C ONCLUSION tributed Coordination Function, IEEE journal on selected areas
We have presented in this paper some simulation in communications, vol. 18, n ◦ 3, March 2000.
results under NS-2 with the ad-hoc routing protocol [9] J. Zhang, L. Cheng, I. Marsic: Models for non-intrusive es-
timation of wireless link bandwidth, Springer-Verlag Lecture
OLSR. They show the effect of radio interferences Notes in Computer Science (LNCS) for Personal Wireless
on the bandwidth obtained by each flow. Because the Communication 2003, vol. 2775, pp.334-348, 2003.
bandwidth is an important quality of service parameter, [10] A.B. Downey: Using pathchar to estimate internet link charac-
teristics, Proceedings of the ACM SIGCOMM’99 Conference
it is necessary to provide the requested bandwidth, as on Applications, Technologies, Architectures and Protocols
long as accepted flows remain active. This is a difficult for Computer Communications, pp.241-250, Cambridge MA,
task if the routing protocol does not take into account August/September 1999.
the radio interference aspect. [11] K. Lai, M. Baker: Measuring link bandwidths using a de-
terministic model of packet delay, Proceedings of the ACM
In this paper we propose to enhance the OLSR routing SIGCOMM 2000 Conference on Applications, Tehnologies,
protocol with bandwidth reservation mechanism. In order Architectures and Protocols for Computer Communications,
to provide a good estimation of the bandwidth needed by pp.283-294, Stockholm, Sweden, August 2000.
a flow, the bandwidth amount that is reserved at nodes [12] V. Paxson: End-to-end internet packet dynamics, IEEE/ACM
Trans. Networking 7, pp.277-292, 1999.
is proportional to the bandwidth requested by that flow
and to its path length.

479

Das könnte Ihnen auch gefallen