DFMAC: DTN-Friendly Medium Access Control
for Wireless Local Area Networks
Hao Liang and Weihua Zhuang
Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, Ontario, Canada, N2L 3G1
Email: [email protected], [email protected]
Abstract—In this paper, we consider a wireless communication
network for both low-speed mobile nodes in densely populated
hotspots and high-speed mobile nodes roaming in a large
area. Wireless local area networks (WLANs) are deployed in
the hotspots, while a delay/disruption tolerant network (DTN)
provides services to roaming nodes in the large area with low
node density. We investigate radio resource allocation for a
DTN/WLAN integrated network, and propose a DTN-friendly
medium access control (DFMAC) scheme for the hotspots. Analytical models are established to characterize the interactions
between a DTN and WLANs under the proposed DFMAC.
Numerical and simulation results demonstrate that our proposed
MAC scheme can provide an effective and efficient way to balance
the radio resource allocation between a DTN and WLANs.
I. I NTRODUCTION
Nowadays, wireless local area networks (WLANs) are
widely deployed at communication hotspots such as libraries,
cafeterias, classrooms, and other public areas, because of
their inherent low implementation cost [1]. To further enlarge
wireless service coverage, the notion of wireless metropolitan
area sharing networks (WMSNs) can be employed [2], where
publicly and/or privately owned wireless Internet access points
(APs) are shared. However, the main problem in WMSNs is
that the deployment of the existing APs limits the service
coverage area and hence caps system performance. Without
sufficient APs, network connections for a mobile node become
intermittent when the node roams outside the hotspots. In fact,
this problem exists in other network environments such as
wireless campus area networks (WCANs) [3] where WLANs
are constrained in separate buildings, and rural area networks
(RANETs) [4] where hotspots are separated by large physical
distances.
As a widely accepted solution for intermittent network
connections, a delay/disruption tolerant network (DTN) can
provide robustness for data transmissions [5]. The key idea
behind DTNs is the store-carry-forward routing [6], where
data messages are stored and carried by wireless nodes for a
considerably long period of time. Whenever a communication
opportunity arises, the stored and carried data messages can
be delivered.
In the literature, most of the existing work on DTN routing
assumes a sparse node distribution and a low traffic load,
whereby the issue of link-layer packet collisions is not taken
into account. However, a recent study indicates that the linklayer contention problem cannot be ignored, especially in
densely populated areas (e.g., small communities) [7]. How
to deal with link-layer contention within the DTN framework
is a challenging yet open issue. On the other hand, there
exist extensive research results for medium access control
(MAC) in WLANs [8]. The recently proposed token-based
method in [9] not only achieves high channel utilization but
also provides service differentiation for best-effort data traffic
and delay sensitive voice traffic. However, the existing work
focuses on nodes in hotspots only. How to acquire effective
and efficient MAC in the presence of nomadic nodes needs
further investigation.
In this work, we consider a DTN/WLAN interworking
environment. Different from other interworking paradigms
such as celluar/WLAN integrated networks [1] [10], in the
DTN/WLAN interworking environment under consideration,
ubiquitous wireless coverage is not provided or not even
available to nomadic nodes. Nomadic nodes can only get
real-time Internet access within the hotspots equipped with
WLANs. We focus on the resource allocation within hotspots
when both nomadic nodes and local nodes are present. In order
to effectively and efficiently balance radio resource allocation
between local traffic and delay tolerant data traffic, we propose
a DTN-friendly MAC (DFMAC) scheme for WLANs based
on a two-phase token passing strategy. By tuning the phase
duration parameters, the service performance tradeoff for
local nodes and nomadic nodes can be achieved. Analytical
results are provided to evaluate the performance of DFMAC.
We further validate our performance analysis with extensive
computer simulations. To the best of our knowledge, this is
the first work in the literature to capture the interworking of
DTN/WLAN and devise a strategy to balance radio resource
allocation between these two network components.
The remainder of the paper is organized as follows. In
Section II, we present the system model under consideration.
In Section III, the proposed DFMAC scheme is discussed in
details. The performance analysis of DFMAC is presented in
Section IV, followed by numerical results in Section V. We
conclude this research in Section VI.
II. S YSTEM M ODEL
Consider a DTN/WLAN integrated network as shown in
Fig. 1, where a number of small hotspots (Hk , k = 1, 2) are
scattered around a large network region. Wireless Internet APs
attached to the Internet backbone provide wireless access to
the nodes in their hotspots. Under this network architecture,
we consider two kinds of wireless nodes, namely 1) the local
Beacon
Beacon
T
Internet Backbone
Dedicated
Phase
T
Dedicated
Phase
Sharing Phase
Sharing Phase
Time
TB
TB
TD(k)
UL Subphase DL Subphase
AP
H1
TD(k)
Local Node
TUL(k)
TDL(k)
Fig. 2: Synchronous mode.
Nomadic Node
H2
Fig. 1: An illustration of the DTN/WLAN integrated
network.
nodes residing in hotspots and 2) the nomadic nodes which can
roam within the entire network region. All mobile nodes and
APs are equipped with a short range radio transceiver. Two
nodes (including APs) can communication with each other
only when they come into the transmission range of each
other. We assume the wireless network within a hotspot is fully
connected and the connections are reliable, but the network
connections become intermittent for nodes outside hotspots.
In this work, nomadic nodes comprise a DTN, whereas local
nodes in a hotspot comprise a WLAN. The nomadic nodes
can stay disconnected for hours and travel through a hotspot
within several minutes. With unique features of a DTN and
a WLAN, the notion of a DTN/WLAN integrated network is
imperative in order to better serve nomadic nodes when they
travel within hotspots.
In the DTN/WLAN integrated network, two kinds of traffic
should be differentiated in resource allocation of a WLAN,
i.e., local traffic and delay tolerant data traffic. The local traffic
is generated by local nodes and destined for a server in the
Internet (uplink) or vice versa (downlink), while delay tolerant
data traffic is generated by nomadic nodes and destined for a
server in the Internet (uplink) or vice versa (downlink). In
the DTN framework, traffic is represented in self-contained
messages (referred to as bundles [5]). Segmentation of a
message into packets is necessary for link-layer transmissions
in a WLAN.
The two types of traffic have different QoS requirements.
For local traffic, the throughput of data service and packet delay of delay-sensitive service are the major QoS measures. On
the other hand, for delay tolerant data traffic, low throughput
and large delay are acceptable. However, as storage capability
is usually constrained for mobile nodes, delivery probability
under a limited buffer space becomes the most important
performance metric. In particular, as the time that a nomadic
node remains in a hotspot is expected to be very brief, its
messages should be transmitted with high priority when the
node is connected via WLAN. In other words, nomadic nodes
are assigned higher priority over local nodes in packet trans-
mission. To provide effective traffic differentiation, resource
reservation for higher-priority nomadic nodes is indispensable
[11], to be discussed in Section III-A. In this work, we
consider the case where messages are delivered by direct
transmissions between nomadic nodes and the APs, while the
impact of message relaying between nomadic nodes is left for
further work.
III. DTN-F RIENDLY M EDIUM ACCESS C ONTROL
(DFMAC)
The DFMAC is a token-based scheme designed for WLANs
where local and nomadic nodes coexist. We consider two operation modes: synchronous mode and asynchronous mode. The
synchronous mode is triggered by APs at hotspots when both
local and nomadic nodes are present, while the asynchronous
mode functions when only local nodes exist.
A. Synchronous Mode and Asynchronous Mode
As shown in Fig. 2, under the synchronous mode, time is
partitioned into superframes of constant duration T . At the
beginning of each superframe, there is a short beacon period
TB for synchronization, followed by a dedicated phase and a
sharing phase. These two phases are dedicated to nomadic
nodes and local nodes, respectively. Each dedicated phase
is further divided into an uplink (UL) subphase for uplink
delay tolerant data traffic and a downlink (DL) subphase
for downlink delay tolerant data traffic. The UL and DL
subphases can have a maximum fraction, βU L (k) and βDL (k),
respectively, of T , where k is the index of a hotspot. Under
non-preemptive service discipline, the maximum durations
of UL and DL subphases within each superframe, TU L (k)
and TDL (k), can be calculated. The maximum duration of a
dedicated phase is given by TD (k) = TU L (k) + TDL (k). The
remaining period in the superframe is the sharing phase.
There is a token in each WLAN. Under the synchronous
mode, during each packet transmission period, only the token
holder can transmit a packet. The token is circulated among
nomadic nodes and the AP in the dedicated phase, and among
local nodes and the AP in the sharing phase. Only nomadic
nodes can access the UL subphase and only the AP can access
the DL subphase. Since all downlink delay tolerant data traffic
queues reside at the Internet and are transmitted via the AP,
when the AP holds the token in a DL subphase, the token is
circulated among these queues.
In the asynchronous mode, there is no superframe structure,
and all local nodes contend for the wireless channel in a
distributed manner.
By introducing the synchronous mode, radio resources can
be reserved for nomadic nodes since the time for them to
travel through a hotspot is in general very brief. But the
DFMAC differs from previous superframe-based methods with
fixed superframe structures [12] [13]. For the DFMAC, the
synchronous mode is only triggered when nomadic nodes are
in the hotspot. Therefore, the overhead introduced by our
superframe structure can be reduced when there is no nomadic
node in the hotspot.
queues, where the AP always holds the token. The truncated
token passing procedures for the UL and DL subphases are
restarted after the beacon period of each superframe.
Obviously, when the message arrival rate of delay tolerant
data traffic is low, the probability for a message to arrive during
a dedicated phase is low accordingly since the duration of each
dedicated phase is short. By using truncated token passing
policy, unnecessary token passing within each dedicated phase
can be reduced when uplink or downlink delay tolerant data
traffic queue is empty, and the unutilized proportion of the
superframe can be passed on to the sharing phase for better
performance of local traffic.
B. Intra-Phase Token Passing
The framework of probabilistic token passing proposed
in [9] is adopted for intra-phase token passing, due to its
good performance and its capability in supporting service
differentiation. For a group G of nodes (or queues) contending
for wireless channel access, the token passing scheme can
be described by two parameters: p = {pi , 1 ≤ i ≤ |G|}
and P = {Pi,j , 1 ≤ i, j ≤ |G|}, where pi denotes the token
holding probability of node i, Pi,j represents the token passing
probability from node i to node j, and |G| is the size of G.
In this work, without service differentiation, the values of pi
and Pi,j are given by
C. Inter-Phase Token Passing
pi = 1/|G|
Pi,j
(
1/(|G| − 1),
=
0,
(1)
if i 6= j
otherwise.
(2)
Specifically, when |G| = 1, the only node (e.g., node i) will
always hold the token by letting pi = Pi,i = 1.
If a token holder has a packet to transmit, the token is
piggybacked to the packet; otherwise, it simply passes the
token to the next node. Before each transmission, there is a
fixed period of time τI for the current token holder to wait
after the wireless channel becomes idle.
Both the asynchronous mode and the sharing phase of
the synchronous mode use the above token passing process.
For the dedicated phase, a truncated token passing policy is
adopted. If the queue of a certain nomadic node is drained, the
token will not be passed onto that node any more within the
current superframe. In the UL subphase, suppose a set GU L of
nomadic nodes in a hotspot are contending for channel access.
If node m, m ∈ GU L , drains off its queue, it broadcasts a
DRAINED message to all other nodes and passes the token to
one of them with equal probabilities. For the remaining nodes,
upon receiving the DRAINED message, they recalculate the
token passing probability by replacing GU L with GU L \ {m}.
The above procedure continues until there is only one
nomadic node left in the UL subphase and its queue is also
drained or the boundary of UL subphase is reached or crossed.
In such cases, inter-phase token passing is initiated to pass the
token to the DL subphase, to be discussed in Section III-C.
The truncated token passing policy is virtually adopted in
the DL subphase for different downlink delay tolerant data
Inter-phase token passing is required only in the synchronous mode where there are multiple phases/subphases. For
inter-phase token passing, each node (including the AP) keeps
three synchronized timers to track how much time has elapsed
within the current superframe, UL subphase, and DL subphase.
These timers are restarted accordingly at the beginning of each
superframe, UL subphase, and DL subphase.
In the dedicated phase, before each packet transmission, the
token holder checks the synchronized timer and compares it
with the boundary of each subphase, i.e., TU L (k) and TDL (k)
in hotspot k. After the current transmission is finished, if the
boundary is reached or crossed, the next token holder is set
to a node (or a queue) belonging to the next phase with the
same probability as the token holding probability of the target
node in that phase.
As data transmissions in the beacon period are not allowed,
if the time left in the sharing phase within the superframe
is not sufficient for transmitting one more packet after the
current transmission, the current token holder passes the token
through the current transmission to a node in GU L with the
same probability as the token holding probability of the node
in the subphase. However, the new token holder will not start
its transmissions until the beacon period of the next superframe
is finished.
D. Node Joining and Leaving
When a node joins or leaves the WLAN, it first waits
for a short period of time τJL (τJL < τI ) after sensing an
idle channel and sends a JOIN or LEAVE message. After
receiving the message, all other nodes within the WLAN will
stop transmitting and the current token holder will discard the
token. Then the AP recalculates token passing parameters (pp
and P ) for all the phase and subphases. If the operation mode
is to be switched from asynchronous mode to synchronous
mode because of the joining of a nomadic node, the subphase
duration parameters TU L (k) and TDL (k) should also be
calculated. Then the AP will broadcast a NOTE message to
inform all nodes of the updated information as well as the
operation mode and a node to which a new token is issued.
If the asynchronous mode is to start, a new token passing
procedure will begin after τI from a specified node. Otherwise,
the synchronous mode will begin after τI with the beacon
signal from the AP, followed by the UL subphase starting
with the specified new token holder.
E. Failure Recovery
The AP is in charge of the failure recovery. It monitors the
current token passing procedure. If a new transmission does
not begin from the current token holder i after τF , where τF
is slightly larger than τI , a new token will be issued by the AP
and passed to node i again. This procedure will continue for
at most NF times. After that, if there is still no transmission
initiated by node i, the node is considered to be inactive in the
system due to battery depletion or hardware impairments. A
NOTE message is then broadcasted by the AP by considering
node i as a leaving node. (Note that the node mobility is
not considered as a cause of failure since the node leaving
procedure can be applied by a node when it tends to move out
of a hotspot as discussed in Section III-D.) During the recovery
procedure, the passing of a new token by the AP is considered
the same as a normal token passing. If the inter-phase token
passing condition is satisfied according to the discussion given
in Section III-C, the AP will stop the recovery procedure and,
at the same time, generates a new token and passes it to a
node belonging to the new phase accordingly.
IV. P ERFORMANCE A NALYSIS
For analytical tractability, the following assumptions are
made:
1) All mobile nodes and APs are equipped with the same
radio transceivers with a transmission range RT R . The
transmission range is much smaller compared with the
dimension of network coverage area α. Each hotspot
covers a circular region with diameter φ. There is no
node failure.
2) The movement of nomadic nodes follows a random
direction (RD) mobility model [14], and the density of
nomadic nodes is low.
3) Data traffic is considered for both nomadic nodes and
local nodes. Both packet arrival of local traffic and
message arrival of delay tolerant data traffic are independent and follow a Poisson process. Each message is
composed of a constant number of K packets.
For a nomadic node, define inter-visiting time as the duration between its two consecutive visits to a hotspot, and
sojourn duration as the period that it stays within a hotspot
in each visit. Based on assumptions 1) and 2), the intervisiting time is approximately exponentially distributed [14].
We denote the rate that a nomadic node visits a hotspot (intervisiting rate) by γV , which is the reciprocal of inter-visiting
time. Although the distribution of sojourn duration is not
obtainable, its expectation TSJ can be derived as a special
case of the average contact duration between two mobile nodes
by letting one of them stationary [14]. The probability that
there are two or more nomadic nodes simultaneously within a
hotspot is negligible with a low nomadic node density.
The average packet arrival rate of local node i within hotspot
Hk is given by λi,k (0 ≤ i ≤ ML,k , 1 ≤ k ≤ MH ), where
ML,k denotes the number of local nodes within hotspot Hk ,
MH is the number of hotspots, and λ0,k represents the local
message arrival rate at the AP of hotspot Hk and destined for
local nodes. For nomadic nodes, the average message arrival
rates of uplink and downlink delay tolerant data traffic are
denoted by λU,i and λD,i (1 ≤ i ≤ MN ), respectively, where
MN is the number of nomadic nodes. The maximum queue
length for uplink delay tolerant data traffic at nomadic node
i is BU,i , and the maximum queue length for downlink delay
tolerant data traffic at a server in the Internet for nomadic node
i is BD,i .
A. Delivery Probability of Delay Tolerant Data Traffic
Since the inter-visiting time is exponentially distributed and
is expected to be much larger than the sojourn duration under
assumptions 1) and 2), the queueing behaviors of both uplink
and downlink delay tolerant data traffic can be described by
continuous time Markov chains (CTMCs). Without loss of
generality, consider the uplink delay tolerant data traffic. For
a tagged nomadic node with traffic arrival parameter λU and
mobility parameters γV and TSJ , define a CTMC XU (t),
where XU (t) ∈ [0, BU ] denotes the number of delay tolerant
messages in the uplink queue and BU is the maximum queue
U
length. The transition rate qi,j
(0 ≤ i, j ≤ BU ) from state i to
state j of the uplink queue can be calculated as follows.
For an uplink message arrival, the arrival rate is given by
λU , thus
U
qi,i+1
= λU , 0 ≤ i ≤ BU − 1.
(3)
When the tagged nomadic node visits one of the hotspots,
messages can be transmitted to the destination via the AP. If
after the current visit, the uplink delay tolerant traffic queue
is still not empty, the transition rate is given by
U
qi,j
=
MH
X
γV I(SU L (k), i − j), 1 < i ≤ BU , 1 ≤ j < i
(4)
k=1
where I(i, j) is an indication function which equals 1 if i is
equal to j, and 0 otherwise. The function SU L (k) denotes the
maximum number of uplink delay tolerant messages that can
be transmitted during the sojourn duration when hotspot k is
visited and is given by
TSJ TU L (k)
(5)
SU L (k) = round
T
TP K
and TP is the time needed for transmitting one packet.
Note that we have made an approximation on the fractional
superframe in the calculation of SU L (k) by considering the
fact that the duration of a superframe is much shorter than the
sojourn duration. For the same reason, the time spent on the
joining and leaving procedures of a nomadic node is ignored.
On the other hand, if after the current visit, the uplink delay
tolerant traffic queue is empty, the transition rate is given by
U
qi,0
=
MH
X
k=1
γV C(SU L (k), i), 1 ≤ i ≤ BU
(6)
where C(i, j) is a comparison function which equals 1 if i
is greater than or equal to j, and 0 otherwise. For any other
state transitions not mentioned precedingly, the transition rate
is equal to 0.
Obviously, the CTMC is a finite state irreducible Markov
chain, and therefore it is ergodic and the stationary probability exists. Denote
the stationary probability by π U =
πiU , 0 ≤ i ≤ BU . It can be calculated by solving a set of
balance equations [15].
The delivery probability of uplink delay tolerant data traffic
U
is given by PD,U L = 1 − πB
U
obtain
. Similarly, we can
the stationary probability π D = πiD , 0 ≤ i ≤ BD and the
delivery probability PD,DL for the downlink.
B. Throughput of Local Traffic
The main system performance measure for data traffic is
normalized system throughput, which is defined as the fraction
of channel time occupied by packet payload transmissions. In
this work, we consider the worst case throughput for local
traffic when the dedicated phase is fully occupied by nomadic
nodes. Note that under truncated token passing policy described in Section III, the actual normalized system throughput
may be better than the worst case if the message arrival rate
of delay tolerant data traffic is low.
Let TR,i,k be the token rotation time of local node i
within hotspot Hk , which is the time duration between two
consecutive token holdings of this node. Then, the queue
utilization of node i within hotspot Hk can be calculated by
ρi,k = λi,k E [TR,i,k ] .
(7)
For the synchronous mode, define a random variable Di,k ,
where Di,k = 1 when the dedicated phase is included in the
token rotation cycle of node i within hotspot Hk and Di,k = 0
otherwise. Therefore, the average token rotation time is given
by
E [TR,i,k ]
= E [TR,i,k |Di,k = 0] P (Di,k = 0)
+E [TR,i,k |Di,k = 1] P (Di,k = 1),
0 ≤ i ≤ ML,k , 1 ≤ k ≤ MH
(8)
where
and
P (Di,k = 0) = 1 − P (Di,k = 1).
(12)
By solving the set of equations given by (8), we obtain
the queue utilizations ρi,k (0 ≤ i ≤ ML,k , 1 ≤ k ≤ MH ).
Then the normalized system throughput of local traffic within
hotspot Hk under synchronous mode can be calculated as
PML,k
T − TB − TD (k)
TP L ρi,k
T hk =
· PML,k i=0
(13)
T
i=0 [TP ρi,k + TT (1 − ρi,k )]
where TP L is the time needed for packet payload transmission.
For the asynchronous mode, we can recalculate the set of
equations (8) by letting TB = TD (k) = P (Di,k = 1) = 0 and
obtain another set of queue utilizations ρ∗i,k (0 ≤ i ≤ ML,k ,
1 ≤ k ≤ MH ). Then, the normalized system throughput is
given by
PML,k
∗
i=0 TP L ρi,k
∗
i.
(14)
T hk = PM h
L,k
TP ρ∗i,k + TT (1 − ρ∗i,k )
i=0
V. N UMERICAL AND S IMULATION R ESULTS
In this section, numerical and simulation results are presented to demonstrate the performance of DFMAC. According
to [9], the time for packet transmission, token passing and
packet payload transmission are given by TP = 0.9793 ms,
TT = 0.3960 ms and TP L = 0.7273 ms. Each message
consists of K = 100 packets. There are two non-overlapping
hotspots (MH = 2) with arbitrary deployment locations in
the network region. The tagged nomadic node is considered
as a pedestrian with RD mobility speed 1.5 m/s and radio
transmission range RT R = 250 m. The diameter of each
hotspot is φ = RT R . Using typical RD mobility parameters
[14], for network coverage area α = 2000 × 2000 m2 , we
have γV = 9.1912 × 10−5 s−1 , TSJ = 107.1038 s; for
α = 2500 × 2500 m2 , we have γV = 5.8824 × 10−5 s−1 ,
TSJ = 107.1038 s. Note that the same sojourn duration is
obtained since it only depends on the RD mobility parameters
and is not affected by the size of network. Furthermore, the
length of each superframe is set to T = 150 ms, and the
duration of beacon period is TB = 1 ms.
ML,k
E [TR,i,k |Di,k = 0] =
X
n=0
[TP ρn,k + TT (1 − ρn,k )]
(9)
A. Delivery Probability of Delay Tolerant Data Traffic
Consider the uplink transmission. The delivery probability
is evaluated in four cases with different configurations of
E [TR,i,k |Di,k = 1] = E [TR,i,k |Di,k = 0] + TB + TD (k) (10) network coverage area (α), uplink delay tolerant data traffic
and TT is the time needed for token passing. Since there arrival rate (λU ), and buffer size (BU ), given as follows.
2
is only one token rotation cycle with Di,k = 1 in each Case 1: α = 2000 × 2000 m , λU = 0.03 message/s,
B
=
800
messages;
Case
2:
α = 2000 × 2000 m2 ,
U
superframe, the time occupied by token rotation cycles with
λ
=
0.03
message/s,
B
=
300 messages; Case 3:
U
U
Di,k = 0 in each superframe is T − E [TR,i,k |Di,k = 1]. Then
2
α
=
2500
×
2500
m
,
λ
=
0.03
message/s, BU =
U
we have
300
messages;
Case
4:
α
=
2500
×
2500 m2 , λU =
1
P (Di,k = 1) =
0.06
message/s,
B
=
300
messages.
For illustration
U
(T −E[TR,i,k |Di,k =1])
1 + E[TR,i,k
|Di,k =0]
clarity, both hotspots H1 and H2 have the same maximum
E [TR,i,k |Di,k = 0]
fractions for UL subphases, i.e., βU L (1) = βU L (2) = βU L .
=
(11)
Fig. 3 shows the message delivery probability of uplink delay
T − TB − TD (k)
0.8
1
0.7
0.9
Normalized system throughput
Delivery probability
0.8
0.7
0.6
0.5
0.4
0.3
0.2
Case 1 (analysis)
Case 2 (analysis)
Case 3 (analysis)
Case 4 (analysis)
0.1
0
0.05
0.1
0.15
β
Case 1 (simulation)
Case 2 (simulation)
Case 3 (simulation)
Case 4 (simulation)
0.2
0.25
0.6
0.5
0.4
0.3
Asynchronous mode (analysis)
Synchronous mode, βUL=0.1 (analysis)
0.2
Synchronous mode, βUL=0.2 (analysis)
0.1
Asynchronous mode (simulation)
Synchronous mode, βUL=0.1 (simulation)
Synchronous mode, βUL=0.2 (simulation)
0
0.3
0
20
40
UL
Fig. 3: The delivery probability of uplink delay tolerant data
traffic versus βU L .
Delivery probability
UL
0.7
= 0.20
0.8
βUL = 0.15
0.7
βUL = 0.10
Normalized system throughput
β
βUL = 0.05
0.6
0.5
0.4
0.3
0.6
0.5
0.4
0.3
Asynchronous mode (analysis)
Synchronous mode, βUL=0.1 (analysis)
0.2
Synchronous mode, βUL=0.2 (analysis)
0.1
Asynchronous mode (simulation)
Synchronous mode, βUL=0.1 (simulation)
0.2
0.1
0
140
0.8
βUL = 0.25
0.9
120
Fig. 5: The normalized system throughput versus packet
arrival rate (10 local nodes).
βUL = 0.3
1
60
80
100
Packet arrival rate (packet/s)
Synchronous mode, βUL=0.2 (simulation)
0
0.02
0.04
0.06
Message arrival rate (message/s)
0.08
0.1
Fig. 4: The delivery probability of uplink delay tolerant data
traffic versus message arrival rate.
tolerant data traffic versus βU L . It can be observed that, the
delivery probability increases as βU L increases, since more
radio resources are provided to a nomadic node when it comes
into a hotspot. The delivery probability can also be improved
with a larger buffer size or a lower message arrival rate because
of the lower message blocking probability in the uplink queue.
On the other hand, the network area has a negative effect
on the delivery probability. For a given set of RD mobility
parameters, the sojourn duration is independent of the network
area, but the inter-visiting time becomes longer for a network
with a larger area, resulting in a lower delivery probability. It
is observed that the analytical and simulation results closely
match with each other.
In order to evaluate the effectiveness of DFMAC under
different uplink delay tolerant data traffic arrival rates, we fix
the value of α and BU to 2500×2500 m2 and 300 messages,
respectively. From the results in Fig. 4 for different βU L
values, we can see that, with an increase of message arrival
rate, the delivery probability of uplink delay tolerant data
traffic decreases. As expected, the delivery probability can be
increased by choosing a larger βU L value. Nonetheless, for a
fixed message arrival rate, the degree of the message delivery
improvement dwindles when the value of βU L increases. The
rationale is that, as βU L increases, the limited buffer size
gradually becomes the bottleneck of message delivery.
0
0
20
40
60
80
100
Packet arrival rate (packet/s)
120
140
Fig. 6: The normalized system throughput versus packet
arrival rate (20 local nodes).
B. Throughput of Local Traffic
To evaluate local traffic throughput, we take hotspot H1 as
an example. For illustration clarity, the same value of packet
arrival rate is chosen for all local nodes and the AP. Since
the influence of βU L and βDL (βDL (1) = βDL (2) = βDL )
on local throughput is equivalent, we set βDL to 0.1 and tune
the value of βU L to show its effect. The results for 10 local
nodes (ML,1 = 10) and 20 local nodes (ML,1 = 20) are
shown in Fig. 5 and Fig. 6, respectively. We can see that,
the analytical and simulation results match well with each
other. As the packet arrival rate of each node increases, the
normalized system throughput increases almost linearly before
saturation, and the speed of increment is much higher for the
case of 20 local nodes. For synchronous mode with a larger
value of βU L , the saturation occurs at a smaller packet arrival
rate. But the normalized saturated throughput does not change
much with the number of local nodes for a given βU L value,
since the sharing phases are already fully occupied by packet
transmissions no matter how many local nodes are present.
This phenomenon can also be explained by (13): As all ρi,k
approaches 1, the second factor on the right hand side remains
unchanged even if the value of ML,k increases. As a result,
the normalized system throughput depends only on the first
factor which is a non-increasing function of βU L .
Sync. mode, 70 packet/s
Sync. mode, 60 packet/s
Sync. mode, 50 packet/s
Sync. mode, 40 packet/s
Async. mode, 70 packet/s
Async. mode, 60 packet/s
Async. mode, 50 packet/s
Async. mode, 40 packet/s
Normalized system throughput
0.55
0.5
0.45
0.4
0.35
0.3
0.25
0.05
0.1
0.15
0.2
0.25
0.3
βUL
0.35
0.4
0.45
0.5
0.55
Fig. 7: The normalized system throughput versus βU L .
The effect of βU L on the normalized system throughput is
further demonstrated in Fig. 7, where 10 nodes are located in
the hotspot with four local traffic arrival rates (packet/s) of
each local node (including the AP). Note that this set of local
traffic arrival rates do not result in throughput saturation in
the asynchronous mode as shown in Fig. 5. From Fig. 7, we
observe that, as βU L increases, the normalized system throughput in the synchronous mode is almost the same as that in the
asynchronous mode when βU L is small. This is because, when
the queue utilization of local nodes is low, the probability for
local queues to be empty is high, and a considerable proportion
of channel time is occupied by pure token passing. On the
other hand, as βU L increases, the queue utilization of local
nodes increases and the overhead on the pure token passing
decreases; therefore the impact of βU L on the normalized
system throughput is not obvious. However, when the βU L
value increases, the normalized system throughput degrades
dramatically due to throughput saturation in the sharing phase.
Under a saturation condition, the sharing phases are fully
occupied by packet transmissions, therefore the normalized
system throughput only depends on the duration of sharing
phases. With nomadic nodes in hotspots, the only possibility
to facilitate delay tolerant data traffic is to sacrifice local traffic
throughput. A theoretical explanation can also be given by
(13), since the normalized system throughput depends only
on the first factor on the right hand side as all ρi,k approaches
1. Note that the throughput saturation starts at a higher βU L
value for a lower local traffic load. In fact, our results confirm
that resource reservation and throughput increment are two
conflicting performance metrics [13]. Therefore, procuring a
desired balance between resource reservation and throughput
increment for a DTN/WLAN integrated network is imperative,
yet left for further work.
VI. C ONCLUSIONS AND F URTHER W ORK
In this paper, a DTN/WLAN interworking scenario is considered. A DTN-friendly MAC (DFMAC) scheme is proposed
to facilitate link-layer resource allocation for local nodes and
nomadic nodes in a hotspot. By tuning the phase duration parameters of the DFMAC, the performance balancing between
local traffic and delay tolerant data traffic can be achieved.
The performance of the proposed MAC scheme is evaluated by
analysis and further validated by simulations. Our results show
that, by properly choosing the value of βU L , we can achieve
satisfactory system performance for both nomadic nodes and
local nodes.
Further work includes the performance evaluation of the
DFMAC under different store-carry-forward routing algorithms [6] and an error-prone wireless channel with lost token
recovery [9]. In a DTN/WLAN integrated network supporting
heterogeneous traffic (e.g., voice, video, and data), not only
is service differentiation between nomadic nodes and local
nodes required, but devising an efficient resource allocation
strategy with effective QoS provisioning is also essential. To
acquire an improved system performance, it is indispensable
to optimally balance radio resource allocation between a DTN
and WLANs. How to achieve an optimal performance tradeoff
is of great interest and needs further study.
ACKNOWLEDGMENT
The authors wish to thank Mr. Ho Ting Cheng for his helpful
suggestions which improved the quality of this paper. This
work was supported by a research grant from the Natural Science and Engineering Research Council (NSERC) of Canada.
R EFERENCES
[1] R. B. Ali and S. Pierre, “Optimal Voice Admission Control Performance
under Soft Vertical Handoff in Loosely Coupled 3G/WLAN Networks,”
in Proc. IEEE WCNC’08, pp. 2980–2985, Mar. 2008.
[2] H. Zhu, X. Lin, M. Shi, P. Ho, and X. Shen, “PPAB: a privacy preserving
authentication and billing architecture for metropolitan area sharing
networks,” IEEE Trans. Veh. Technol., to appear.
[3] A. Messier, J. Robinson, and K. Pahlavan, “Performance monitoring of
a wireless campus area network,” in Proc. IEEE LCN’97, pp. 232–238,
Nov. 1997.
[4] M. O. Kachienga, “Technological innovations in health care: Challenges
of managing telehealth technology in South Africa,” in Proc. IEEE
IEMC’08, pp. 1–5, Jun. 2008.
[5] K. Fall, “A delay-tolerant network architecture for challenged internets,”
in Proc. ACM SIGCOMM’03, pp. 27–34, Aug. 2003.
[6] Z. Zhang, “Routing in intermittently connected mobile ad hoc networks
and delay tolerant networks: overview and challenges,” IEEE Commun.
Surveys and Tut., vol. 8, no. 1, pp. 24–37, 2006.
[7] A. Jindal and K. Psounis, “Contention-aware performance analysis of
mobility-assisted routing,” IEEE Trans. Mobile Comput., vol. 8, no. 2,
pp. 145–161, Feb. 2009.
[8] S. Kumar, V. S. Raghavan, and J. Deng. “Medium access control
protocols for ad-hoc wireless networks: a survey,” Elsevier Ad-Hoc
Netw., vol. 4, no. 3, pp. 326–358, May 2006.
[9] P. Wang and W. Zhuang, “A token-based scheduling scheme for WLANs
supporting voice/data traffic and its performance analysis,” IEEE Trans.
Wireless Commun., vol. 7, no. 5, pp. 1708–1718, May 2008.
[10] W. Song and W. Zhuang, “Multi-service load sharing for resource
management in the cellular/WLAN integrated network,” IEEE Trans.
Wireless Commun., vol. 8, no. 2, pp. 725–735, Feb. 2009.
[11] H. T. Cheng and W. Zhuang, “Novel packet-level resource allocation
with effective QoS provisioning for wireless mesh networks,” IEEE
Trans. Wireless Commun., vol. 8, no. 2, pp. 694–700, Feb. 2009.
[12] T. Li, D. Logothetis, and M. Veeraraghavan, “Analysis of a polling
system for telephony traffic with application to wireless LANs,” IEEE
Trans. Wireless Commun., vol. 5, no. 6, pp. 1284–1293, Jun. 2006.
[13] H. T. Cheng, H. Jiang, and W. Zhuang, “Distributed medium access control for wireless mesh networks,” Wireless Commun. Mobile Comput.,
vol. 6, no. 6, pp. 845–864, Sept. 2006.
[14] T. Spyropoulos, A. Jindal, and K. Psounis, “An analytical study of
fundamental mobility properties for encounter-based protocols,” Int’l J.
Auton. Adapt. Commun. Syst. (IJAACS), vol. 1, no. 1–2, pp. 4–40, Jul.
2008.
[15] L. Kleinrock, Queueing Systems, vol. 1. New York: Wiley, 1975.