Sunteți pe pagina 1din 3

IEEE COMMUNICATIONS LETTERS, VOL. 14, NO.

2, FEBRUARY 2010 109

End-to-End Delay in Wireless Random Networks


Seung Min Yu and Seong-Lyun Kim

Abstract—The focus of this letter is to derive a scaling law for


the ene-to-end delay of wireless random networks under node
mobility, where � nodes randomly move with the speed of �. To
that end, we apply the cover time analysis and relate it to the
delay scaling law. As a result, ( √ we
) derive that the mean delay
per S-D pair as Θ(�) or Θ �� , and the worst case delay is
(√ )
� log �
Θ(� log �) or Θ �
, corresponding to one slot time length
1
that is either constant or �√ �
.
Index Terms—Random networks, delay, mobility, cover time.

I. I NTRODUCTION
Fig. 1. Random packet forwarding in a wireless random network.

T HE capacity of the random wireless ad hoc networks has


become an interesting issue since the seminal work by
Gupta and Kumar [1], where they showed
( that
) the throughput Θ(�). However, to the best of our knowledge, there has been
capacity per S-D pair follows Θ √� 1log � , when all uni- no analytic result on the cover time for the mobile node case.
formly located � nodes are stationary. On the other hand, node In this letter, we extend the static cover time to the mobile
mobility can improve the capacity. Grossglauser and Tse [2] case and apply the result to the delay analysis of a wireless
showed that the capacity is independent of the node density, random network with node mobility.
and is quantified by Θ (1), on the condition that each node has Under the condition that each node uses opportunistic
sufficient mobility. However, the capacity of Θ (1) is achieved routing and node mobility satisfies the assumptions of the
with a loose delay constraint, because the speed of mobile next section, our analysis verified previous results that the
movement is very slow compared to the delay requirement. mean delay is Θ (�). Furthermore, we found that Θ (� log �)
Delay under node mobility has been investigated by [3]-[6] constitutes the worst case delay. The delay scaling law we
using different models: Brownian mobility model [3], [4], obtained in this letter applies to both stationary and mobile
random walk mobility model [3], random way-point mobility nodes. In [3], using the critical transmission range, it was
model [5] and i.i.d. mobility model [6]. These studies have shown
(√ that )the mean delay of a stationary network [1] follows

showed that the two-hop relaying scheme in [2] has a mean Θ log � , whereas it was Θ (�) for the mobile case [2].
delay of Θ(�) under different node mobility models. The main reason for such a difference in the result of the
The main purpose of this letter is to revisit delay scaling stationary case is that our analysis assumed opportunistic
laws from the perspective of cover time analysis, a classical routing. On the other hand, an optimal routing path (shortest
research issue in computer science [7]. Applications of cover path) was used for the stationary case in [3]. Delays in
time analysis include searching/querying, routing, membership both stationary and mobile cases were compared in different
services and group-based communications. conditions. That is why we used the condition that each node
The cover time defines the expected number of transmis- used opportunistic routing. Our scaling laws say that the delay
sions until all nodes in the network are visited by a given increases with � in the same order, either in stationary or
packet, where all nodes are stationary and do not have any mobile networks. This means that the node mobility in fact
topology information. Partial cover time is the time required improves the delay performance but cannot change the delay
for a portion of nodes to be visited. In [7], the authors derived scaling law.
that, when the nodes have a critical transmission range [8], the
cover time scales with Θ(� log �) and the partial cover time is II. R ANDOM N ETWORK WITH N ODE M OBILITY
Consider a wireless random network [1] composed of
Manuscript received September 9, 2009. The associate editor coordinating
the review of this letter and approving it for publication was F. Granelli.
uniformly located � nodes that are able to move around the
This research was supported by the MKE (The Ministry of Knowledge area of an unit disc with a constant speed, � meters per second.
Economy), Korea, under the ITRC (Information Technology Research Center) A node having a packet forwards it to a next node selected
support program supervised by the NIPA (National IT Industry Promotion
Agency) (NIPA-2009-C1090-0902-0037).
randomly among nodes within transmission range (Figure 1).
The authors are with the Radio Resource Management & Optimization Lab- This means that the covered node may receive the same packet
oratory, School of Electrical and Electronic Engineering, Yonsei University, and forward it again. The nodes move and transmit their
134 Sinchon-Dong, Seodaemun-Gu, Seoul 120-749, Korea (e-mail: {smyu,
slkim}@ramo.yonsei.ac.kr).
packets in every �(�) time-slot, where �(�) is either constant
1
Digital Object Identifier 10.1109/LCOMM.2010.02.091821 [1], [2] or �(�) = �√ �
[3]. A constant time-slot may miss the
c 2010 IEEE
1089-7798/10$25.00 ⃝

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:53 UTC from IE Xplore. Restricon aply.
110 IEEE COMMUNICATIONS LETTERS, VOL. 14, NO. 2, FEBRUARY 2010

opportunity to forward packets when � is relatively high, but A. Mobile Cover Time in Homogeneous Networks
the latter does not have this problem. Let us define the mobile cover time as the cover time in
In [8], Gupta and√Kumar derived the so called critical a random network, where nodes are moving and forwarding
log �
transmission range �� for guaranteing asymptotic full packets opportunistically as described in Section II. Similarly,
connectivity, with the probability approaching one. As in [3], we define the mobile partial cover time as the duration until
[7], we adopt the same critical transmission range for each the predefined portion of moving nodes are covered. Then, we
node, in order to remove the effect of transmission range in have the following proposition for the homogeneous network:
our investigation on node mobility. We assume the protocol Proposition 1. The mobile cover time in the homogeneous
model [1] for concurrent transmissions. Therefore, multiple network is
packets are transmitted at a given instant. However, without
�ℎ (�) = Θ(� log �) (1)
loss of generality, we focus on the cover time of a single packet
to analyze the average delay performance. This approach was
also used in [3] for delay-throughput tradeoff analysis. Proof: We have:
Let us define that a node is covered by a packet from a
�ℎ (�) = � [�1 + �2 + ⋅ ⋅ ⋅ + �� ]
source if the node receives the packet. From the packet delay
= � [� [�1 + �2 + ⋅ ⋅ ⋅ + �� ∣�1 , �2 , ⋅ ⋅ ⋅ , �� ]]
perspective, the cover time gives the worst case delay for a
packet at the given destination. In that sense, there is a close = � [� [�1 ∣�1 ]] + � [� [�2 ∣�2 ]] + ⋅ ⋅ ⋅ + � [� [�� ∣�� ]]
� [ ] ∑ � �
relationship between cover time and delay. ∑ 1 � ∑ 1
= � = =�
�� is the number of transmissions needed to cover the �-th �=1
� �
�=1
� − (� − 1) �=1

node, given that � − 1 nodes are so far covered by a packet. ≈ �(ln � + �) (2)
�� is the probability that the �-th node is covered by using
one transmission attempt from (� − 1)-th node, given that all where � ≈ 0.5772 is the Euler-Mascheroni constant.
� − 1 nodes are covered. Note that �� is a geometric random Let us consider the mobile partial cover time �ℎ (�, �),
variable if the success probability �� is a fixed value. The when the parameter � (0 < � < 1) defines the portion of
uncovered node ratio �� is a random variable due to random nodes to be covered:
node mobility. We assume that the mean of �� is a function of �ℎ (�, �) = �[�1 + �2 + ⋅ ⋅ ⋅ + ��� ], (3)
�, �, and �. Using simulations with random way-point model
[5], we verify that the mean of �� , denoted by � (�, �, �), where �� is an integer.
increases and converges to a constant as speed � increases for Proposition 2. The mobile partial cover time in the homoge-
given � and �. Due to the space limitation, we do not include neous network is
the simulation results. Let us make the followings for further
�ℎ (�, �) = Θ(�) (4)
steps.
Assumption 1. The node mobility satisfies the condition
that there exists � such that lim � (�, �, �) = �−(�−1)� and Proof: Note that
[  ] �→� ��
lim � �� − �−(�−1)
[ ]
 ≥ � = 0, for any given �. 1
  ∑
�→� � �ℎ (�, �) = � [�1 + �2 + ⋅ ⋅ ⋅ + ��� ] = �
�−(�−1) ��
Definition
[ 1.  If ]� (�, �, �) = � and ⎛ ⎞ �=1
�−(�−1) � (1−�)�
� �� −  ≥ � = 0, for any given �, then the
 

∑ 1 ∑ 1
network is said to be homogeneous. = �⎝ − ⎠
�=1
� �
�=1
Assumption 1 and Definition 1 are related to the Chebyshev
inequality. It is trivially true that the variance of �� decreases ≈ � [(ln � + �) − (ln((1 − �)�) + �)]
( )
to zero as � increases. Therefore, �� is considered to be a 1
= � ln (5)
constant in homogeneous networks. 1−�
Let us imagine covered nodes as molecules of a chemical
solute and the uncovered nodes as molecules of a solvent.
Then, the homogenous condition resembles a homogenous
solution, where the concentrations in any location are equal B. Mobile Cover Time in Non-homogeneous Networks
due to the high speed of molecular movement. A random network is non-homogeneous if it does not satisfy
the homogeneous condition. Then, we have:
Proposition 3. The mobile cover time in the non-homogeneous
III. C OVER T IME A NALYSIS
network is
As mentioned, the cover time defines the worst case delay.
��ℎ (�) = Θ(� log �) (6)
On the other hand, the partial cover time is closely coupled
with the mean delay of the random network. We start with
cover time analysis for the sufficient mobility network which Proof: We have
we defined as the homogeneous network. Then, we apply this � ] ∑[� [ ]
result to the more general mobility of the non-homogeneous
∑1 1
��ℎ (�) = lim � = lim � , (7)
network. �→��ℎ
�=1
�� �=1
�→� �ℎ ��

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:53 UTC from IE Xplore. Restricon aply.
YU and KIM: END-TO-END DELAY IN WIRELESS RANDOM NETWORKS 111

1200 1800

cover ratio =1 (simulation) cover ratio = 1


cover ratio =1 (analysis) 1600 cover ratio = 0.8
cover ratio = 0.8 (simulation) cover ratio = 0.6
1000 cover ratio = 0.8 (analysis) cover ratio = 0.4
cover ratio = 0.6 (simulation) 1400 cover ratio = 0.2
cover ratio = 0.6 (analysis)
cover ratio = 0.4 (simulation)
800 cover ratio = 0.4 (analysis) 1200
cover ratio = 0.2(simulation)
Mobile cover time

Mobile cover time


cover ratio = 0.2 (analysis)
1000
600

800

400 600

400
200

200

0 0
20 40 60 80 100 120 140 160 180 200 20 40 60 80 100 120 140 160 180 200
Number of nodes Number of nodes

Fig. 2. Mobile cover time and mobile partial cover


√ time in the homogeneous Fig. 3. Mobile cover time and mobile partial cover time in the
√ non-
log � log �
case as a function of the number of nodes (� = ��
, � = 0.05). homogeneous case as a function of the number of nodes (� = ,
��
� = 0.01).

where ��ℎ denotes the mobile speed in non-homogenous √


network (0 ⩽ ��ℎ < �). Noting that the mean of �� increases slot. We set the transmission range � = log �
�� for asymptotic
with � according to our background simulations, connectivity of � nodes in the unit disc. Figure 2 contains the
[ ] [ ] [ ] results sampled over 103 instances, which coincide with the
1 1 1 results of (2) and (5). To see the impact of node mobility,
lim � ≤ lim � ≤ lim � (8)
�→� �� �→��ℎ �� �→0 �� we tested the non-homogeneous case with � = 0.01. Figure 3
Therefore, it follows that illustrates those results, where the scaling law is the same as
for the homogeneous case, but the cover time increases on a
�ℎ (�) ≤ ��ℎ (�) ≤ �� ���� (�), (9) constant scale.
where �� ���� (�) denotes the cover time for the stationary
node case. In [7], it was derived that �� ���� (�) = Θ(� log �) IV. C ONCLUSIONS
under the assumption of the critical transmission range of each The mean delay is related to the mobile partial cover time
node. Also, �ℎ (�) = Θ(� log �) from Proposition 1. Thus, and is the product of the mobile partial cover time (in slots)
the mobile cover time under the non-homogeneous condition Θ(�) and the length of one time slot, which is either a constant
is ��ℎ (�) = Θ(� log �). or �(�) = �√1 � . Therefore the mean delay follows Θ(�) or
(√ )
Proposition 4. The mobile partial cover time in the non- Θ �� . Furthermore, the worst case delay is related to the
homogeneous network is
mobile
( √ cover ) time Θ(� log �), which is either Θ(� log �) or
��ℎ (�, �) = Θ(�) (10) � log �
Θ � , depending on �(�).

Proof: It follows that R EFERENCES


[1] P. Gupta and P. R. Kumar, “The capacity of wireless networks,” IEEE
�ℎ (�, �) ≤ ��ℎ (�, �) ≤ �� ���� (�, �), (11) Trans. Inf. Theory, vol. 42, pp. 388-404, Mar. 2000.
[2] M. Grossglauser and D. N. C. Tse, “Mobility increases the capacity of
where �� ���� (�, �) denotes the partial cover time for the ad hoc wireless networks,” IEEE/ACM Trans. Networking, vol. 10, pp.
stationary node case. In [7], it was derived that �ℎ (�, �) = 477-486, Aug. 2002.
[3] A. E. Gamal, J. Mammen, B. Prabhakar, and D. Shah, “Throughput-delay
Θ(�), and from Proposition 2, �ℎ (�) = Θ(�). Thus, trade-off in wireless networks,” in Proc. IEEE INFOCOM, Apr. 2004.
��ℎ (�, �) = Θ(�). [4] X. Lin, G. Sharma, R. Mazumdar, and N. Shroff, “Degenerate delay-
capacity trade-offs in ad hoc networks with Brownian mobility,” IEEE
The results from Propositions 1-4 indicate that the mobile Trans. Inf. Theory and IEEE/ACM Trans. Networking (joint special issue),
cover time follows Θ(� log �), and the mobile partial cover vol. 52, pp. 2777-2784, June 2006.
time follows Θ(�), independent of the mobile speed �. In fact, [5] G. Sharma and R. Mazumdar, “Delay and capacity trade-Off in wireless
ad hoc networks with random way-point mobility,” preprint, 2005.
this scaling law is exactly the same as the cover time for the [6] X. Lin and N. B. Shroff, “The fundamental capacity-delay tradeoff in
stationary case [7]. large mobile ad hoc networks,” in Proc. Third Annual Mediterranean Ad
Hoc Networking Workshop, June 2004.
[7] C. Avin and G. Ercal, “On the cover time of random geometric graphs,”
C. Simulations in Proc. ICALP, July 2005.
[8] P. Gupta and P. R. Kumar, “Critical power for asymptotic connectivity
To verify our analysis, we conducted simulations with in wireless networks,” Stochastic Analysis, Control, Optimization and
sufficient mobility such that the mobile speed is high enough Applications: A Volume in Honor of W. H. Fleming, W. M. McEneany,
G. Yin, and Q. Zhang, Eds. Boston, MA: Birkhauser, pp. 547-566, 1998.
to satisfy the homogenous condition (i.e., �� is considered
to be as a constant). Nodes moved according to the random
way-point model with a maximum speed of 0.05 meters per

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:53 UTC from IE Xplore. Restricon aply.

S-ar putea să vă placă și