2009 International Conference on Emerging Technologies
Multiple End-to-End QoS Metrics Gateway Selection Scheme in Mobile Ad hoc Networks Safdar H. Bouk
Iwao Sasase
Department of Information and Computer Science Graduate School of Science & Technology, Keio University Yokohama, Japan
[email protected]
Department of Information and Computer Science Graduate School of Science & Technology, Keio University Yokohama, Japan
[email protected]
Abstract— Several gateway selection schemes have been proposed that select gateway nodes based on single QoS metric, for instance link availability, link capacity etc. or multiple non-QoS metrics, such as the combination of gateway node speed, residual energy, and number of hops, for Mobile Ad hoc NETworks (MANETs). Each scheme just focuses to improve only single network performance such as network throughput, packet delivery ratio, end-to-end delay or packet drop ratio. However, none of those schemes improves overall network performance. To improve the overall network performance, it is necessary to select a gateway with stable connection, maximum link capacity and less latency. In this paper, we propose a gateway selection scheme that considers multiple end-to-end QoS parameters such as route availability period, route capacity and route latency, to select a potential gateway node. The precise estimation and propagation of end-to-end QoS metrics are proposed in this paper. We simulate our gateway selection scheme with a simple hybrid gateway discovery algorithm and it shows that our gateway selection scheme improves throughput, packet delivery ratio with less energy consumption per node. Also it improves the end-toend delay compared to the single QoS gateway selection schemes. Keywords- gateway selection; QoS Metrics; ad hoc network
I.
INTRODUCTION
Mobile Ad hoc NETworks (MANETs) [1] are on demand, spontaneous, self-administrative, with no infrastructure and plug-n-play type future networks. There are two major MANET scenarios, first one is standalone MANET and the second is interconnected MANET [2]. In standalone scenario, ad hoc network is not connected with any other network and all traffic is destined within the network premises. On the other hand in interconnected MANET, the ad hoc network is connected with infrastructure network(s). Recently, research in the field of MANETs has been aimed on the assimilation of MANETs with the infrastructure networks e.g. cellular [3] or data networks [4] to extend the network services and coverage to network users beyond the bounds of time and place with definite level of Quality of Service (QoS). The communication between infrastructure network and MANETs is provided by some MANET nodes called gateway nodes that are equipped with multiple interfaces. These gateway nodes provide bridge between multiple networks and may be mobile or fixed, as shown in Fig.1. An ad hoc node must discover and select a suitable
978-1-4244-5632-1/09/$26.00 ©2009 IEEE
446
gateway node from multiple gateways before starting communication with the node in the infrastructure network. Hence, the gateway discovery and selection is an important factor to enable integration between both networks. This research lies in the category of the gateway selection.. Gateway selection is a process that selects a potential gateway node out of multiple discovered gateway nodes based on network, link, route or gateway node parameters. In literature, several gateway selection methods [5]-[12] have been proposed that consider different QoS or non-QoS parameters to select a potential gateway node. Most of the gateway selection methods consider hop count, delay, mobility traces, link connectivity and residual load capacity of gateway node or combination of these parameters. The gateway selection schemes in [5] and [6] select prospective gateway based on hop count. A gateway discovery message is broadcasted by the gateway and based on that message each node calculate its distance to the gateway. The gateway with shortest path in terms of hop count is selected for relaying traffic from MANET to the infrastructure network. In [7], Ghassemian et al. proposed a gateway selection scheme that considers delay, jitter, hop count and bit error rate as rate as additive cost of a route between MANET node to gateway router and edge gateway router. A gateway node with a route with minimum cost value is selected by the MANET node for internet traffic. Another gateway selection scheme that considers MobilityTracing-Value (MTV) as a basic criterion to select gateway is proposed in [8]. If a neighboring node does not receive a Hello message until its duration expires, then the MTV value increases. Hence, larger value of MTV denotes the higher probability of link failure. A gateway node on a route with minimum MTV is selected. In case if two routes have the same MTV value then a route with route life time value is considered to select a gateway. If MTV and life time of multiple routes are similar then hop count is the third option to select a gateway. A hybrid gateway selection criterion is proposed in [9]. The weighted sum is computed based on the Euclidean distance between MANET nodes and mobile gateway and load of a gateway node. Based on this weight value a MANET node selects a gateway node. In [10], gateway selection scheme selects a gateway with maximum residual capacity on a path
with less number of hops. The residual capacity of a gateway node is computed by subtracting the current traffic load of the gateway node from it total capacity. A routing protocol for hybrid wired wireless network is proposed in [11] where MANET is connected with internet through gateway nodes. It computes route expiration time based on the node speed, location and the direction of the movement. A gateway node with maximum route expiration time is selected for traffic relaying from MANET to the infrastructure network. Internet
Based on our knowledge, none of the gateway selection algorithm considers the combination of end-to-end (path from a MANET node to gateway node) QoS parameters i.e. route availability period, capacity of route and latency. In this paper, we consider end-to-end QoS parameters to select a potential gateway node. The modified and precise end-to-end path availability estimation that is based on the one in [11] is proposed in this paper. Along with the route availability period, route capacity and latency are also considered to select the potential gateway node. Our proposed gateway selection algorithm is simulated with the simple proactive gateway discovery scheme. However, our gateway selection algorithm can easily be adapted by the any gateway discovery scheme. Rest of the paper is organized as follows: Section II describes the proposed gateway selection scheme and gateway discovery method. Simulation results are discussed in Section III. Finally, Section IV concludes this paper. II.
Cellular Network Gateway Node Ad-hoc Network Node MANET Figure 1. Interconnected MANET.
In [12], a weight based gateway selection algorithm is proposed. It calculates weights of the gateway nodes by considering residual battery power, speed of gateway node and number hops. The gateway with higher weight is selected. This scheme slightly improves the network throughput, however, end-to-end delay and packet drop ratio depends on the proper selection of the weighting factors, which is quite difficult in dynamic scenarios. The gateway selection schemes discussed above either consider available capacity of a gateway node, number of nodes registered with a gateway, movement trace information, route expiration time or bit error rate. The basic gateway selection criterion in almost every scheme is based on the parameters related with gateway node. In MANET multi-hop routes are established from a MANET node to gateway nodes. Hence, it is possible that multiple routes may have some common nodes through which traffic is relayed from a MANET node to the gateway node(s) or vise versa. These common nodes may create bottleneck to degrade the network throughput. Also, it is observed in [13] that average packet delay decreases and network throughput increases when traffic is forwarded over a route with larger end-to-end connectivity duration. It indicates that the network throughput depends not only on the gateway capacity or solely on the route stability but also on the capacity and stability of the route between MANET node and gateway node. Also, selecting a stable route does not guarantee the throughput improvement, because multiple nodes may select this route to forward traffic and cause congestion. In result, it degrades the throughput or increases the end-to-end delay.
447
PROPOSED GATEWAY SELECTION SCHEME
In this section we discuss the proposed gateway selection algorithm along with selection parameters and discovery process. We believe that performance of the gateway selection algorithm solely depends on these gateway selection parameters that directly affect the QoS that an infrastructure network provides to the MANET. Therefore, we consider multiple QoS parameters in our gateway selection algorithm that provide high QoS to MANET. These parameters compute the end-to-end (route between a MANET node and a gateway node) route availability period, capacity of route and latency. A. Gateway Selection Parameters The detailed description of the gateway discovery parameters are as follows. 1) Route Availability Period: In MANET nodes are moving at random speed and direction that result in a dynamic topology. Consider an example of a Random Walk mobility model [15] where movement of each node is a sequence of random length intervals called epochs during which a node moves in a constant direction T at a constant speed v. In this situation the link availability period between two nodes is varying at different time intervals. And the route availability period between two nodes that are not immediate neighbors of each other, is equal to the minimum link availability period between intermediate nodes in that route. The route availability period, Li, of a route i between MANET node and gateway node indicate the total time that a gateway is accessible by a MANET node through that route. Our route availability period estimation is based on the link connectivity prediction method in [11] where Li represents the minimum link availability period and lu is the link availability period between two neighboring intermediate nodes in a route from a source MANET node (S) to the gateway node (G). Li = min{ lu }, (1) where u = each link between intermediate nodes on route i.
In [16], link connectivity period, lu, of link between node m and n is computed as follows. Let node m and n on route i are in transmission range tr of each other and current position of m and n is (xm, ym) and (xn, yn), respectively. Suppose Tm and Tn are moving directions and vm and vn are the speeds of node m and n, respectively. Then, lu of node m and n is computed as:
(DE JG ) (D 2 J 2 )tr 2 (DG EJ ) 2 D2 J 2
lu where,
D vm cos T m vn cos T n , E xm xn , J vm sin T m vn sin T n , and G ym yn . In (2), it is assumed that the epoch lengths of node m and n are same, which is not accurate in real scenarios. This situation is shown in Fig.2. According to the link estimation time in (2), node m and n are estimated to be in the transmission range of each other till time t2, as movement of node n is shown by dashed line after time t1 in Fig.2. However, epoch length of n (e.g. t1 – t0) is shorter than the epoch length of m (e.g. t2 – t0) and after time t1 node n randomly selects another direction and speed. In this situation, m and n are in transmission range of each other till time period of t1 – t0.
In our proposed scheme, lu is estimated as follows. First, lu is computed as in (1). If lu is greater than epoch length of m (em) then lu = em otherwise, if lu is greater than epoch length of n (en) then lu = en. Conversely, if lu is less than em as well as en then the link connectivity time is same as computed by (2). The overall route availability period is computed in similar way by our proposed scheme as in (1). 2) Residual Route Load Capacity: In multi-hop MANET there can be multiple routes to the gateway node(s). Also, there is possibility that multiple routes may have some common node in the routes between mobile nodes to the gateway nodes, as shown in Fig. 3. If traffic is forwarded through these nodes then the common node(s) in the end-to-end routes are overloaded and results in a bottleneck situation that will result in an increased delay and packet loss. Almost all previous proposals just compute the traffic load of a gateway node and based on that information they select gateways. On the contrary, in our proposed scheme we select gateway nodes accessible through a route with maximum available load capacity. The residual load capacity of a route is the minimum available load capacity at any node, including intermediate nodes and the gateway node, in that route. Suppose the maximum load capacity of a node m is P and the current traffic load handled by m is Om then the residual load capacity, cm, at m is computed as:
cm = P - Om
where,
(3)
s
Om
¦r k j
j
.
j 1
Simulation Area
In (3), Om is the current traffic load on node m that is relaying traffic from s traffic sources and rj and kj denote the average packet arrival rate and average packet size of the traffic from source j, respectively. The overall residual load capacity Ci of route i is computed as:
n
m t2 em Node m at t1
en
m
n
n
n t2
t1 y
m
m t0
e = Epoch Length t0 = Initial Time
Figure 2. Mobility pattern in MANET. GW1
(4)
j = nodes in a route including gateway node. 3) Route Latency: Latency is propagation delay plus processing time of a packet from one node to another node. Latency can either be increased when packet is relayed in a hop-by-hop fashion from sender to the receiver node or when traffic load is high on any node in the route. Latency of route i, Yi, is the additive measurement of latency at each link on the route between the gateway and mobile node.
n
n
x
Ci = min {cj}
where,
en
GW2
as:
In last, the overall QoS value of a route i, Gi, is computed
Gi
Bottleneck
§ Li ¨ © Lmax
· § Ci ¸¨ ¹ © Cmax
· § Ymin · ¸¨ ¸ ¹ © Yi ¹
(5)
where Lmax , Cmax , and Ymin is maximum route availability period, maximum residual route load capacity and minimum router latency from all the available routes between a MANET node and gateway nodes. After computing Gi for every route, a gateway that is accessed through a route with maximum Gi is
Traffic Source
Figure 3. Bottleneck situation in MANET.
448
selected by the MANET node. A user can also set some preferences for individual parameter in Gi to prioritize any of the parameters based on the network preferences. B. Proposed Gateway Selection and Discovery Scheme. In this section, we briefly discuss the gateway selection and discovery algorithm and also the QoS parameter propagation during gateway discovery process. We analyze our gateway selection scheme in the hybrid gateway discovery algorithm [14], where each gateway node periodically advertises its parameters within a proactive region of k-hops by sending Gateway Advertisement (GW_ADV) message and the nodes that are out of the proactive zone discover gateway(s) in a reactive manner by sending Gateway Discovery (GW_DISC) message, as shown in Fig.4. However, our proposed gateway selection scheme can easily be adopted in the reactive or proactive gateway discovery method. Before broadcasting each periodic GW_ADV message, a gateway node calculates its parameters i.e. epoch (eG), speed (vG), direction (TG), location (xG ,yG), time_stamp, capacity cG (Ci :=cG), and assign Li = Yi= 0. In next step, gateway node updates these parameters in GW_ADV message and broadcast in a proactive region of the MANET by using the TTL (time to live) value. When a mobile node receives a GW_ADV message, it computes its own parameters i.e. eu , vu, Tu, (xu ,yu), cu, and cG:=Ci. Then a mobile node, calculates the lu as in (2) and considers the minimum value to the Li out of lu, eu ,and eG. Also, it computes the latency Yi and then it compares its own residual load capacity with the gateway node capacity and minimum capacity is assigned to Ci. In the end, it adds or updates the route entry in the routing table with the computed parameters. And then, it updates these QoS route parameters (i.e. Ci, Li and Yi) in the GW_ADV message and forwards this message. When other MANET nodes receive that GW_ADV message they also perform the same steps as discussed above, update their routing tables and forward this message until TTL value reaches the k-hops. In the GW_ADV message parameters Ci, Li and Yi represent the minimum route capacity, minimum route availability period, and the route latency, respectively. GW 0
GW 1
Reactive Zone
Proactive Z Zone(s)
Figure 4. Hybrid gateway discovery algorithm.
The MANET nodes in the reactive region send GW_DISC message to discover the gateway(s). A MANET node sends
449
GW_DISC with its own parameters i.e. eu, vu, Tu, xu ,yu, time_stamp, and cu. GW_DISC message processed by each node at every hope and the minimum parameters of the route are forwarded until it is received by any gateway node or a any node in the proactive region. If a node in the proactive region receives GW_DISC message, it will send a unicast GW_ADV message to the sender of the GW_DISC message. Before sending a unicast advertisement message, the proactive region node finds the best available path to the gateway node from its routing table and it will compare that route parameters (capacity and availability period) with the one in the GW_DISC message. The minimum of both the parameters along with the sum of route latency in routing table and the latency of the GW_DISC message are added in the unicast advertisement message. On receiving the unicast advertisement message, the mobile node will update its routing table. The MANET is a dynamic topology network where data traffic is generated and forwarded through dynamic routes. In results, the overall route capacity either increases or decreases in random. Therefore, it is necessary to propagate the current state of the route to the data traffic source node(s). The intermediate node on the active route sends route update message to the data traffic source node(s) in a unicast manner when a new connection is established through this route or an old connection is terminated. In this manner a data traffic source node(s) select a potential gateway by using the updated route parameters. In this manner, the route entries of route(s) to the gateway node(s) along with their QoS parameters are maintained in the routing table at each MANET node. If a node wants to send traffic to the host in infrastructure network, it calculates overall QoS value of each route (Gi) in the routing table and select the route with maximum Gi. TABLE I.
TABLE TYPE STYLES
Parameter Area
Value 1000m x 1000m
Wireless MAC Interface
IEEE 802.11b
Frequency
2.472 GHz
Propagation Model
Two Ray Ground
Data Rate
2 Mb
Transmission Range
175m
Number of Nodes
25
Number Gateway Nodes
5
K
2-hops
Mobility Model
Random Walk
Gateway Node Mobility / Speed
1m/s, 2m/s,…, 5 m/s
CBR Packet Size
512 bytes
CBR Interval D
0.3s, 0.15s, ….. , 0.09s
GW_ADV Interval
10s
Time between GW_DISC Messages
5s
Simulation Time
1000s
III.
bandwidth and minimum number of hops. On the other hand, conventional scheme 2 only considers path stability. Due to the node mobility and multiple CBR connections, paths are not stable and path capacity varies in time and neither of the conventional schemes can cope with dynamic nature of the MANET. In result, the CBR success ratio in both the schemes is very less.
SIMULATION PARAMETERS
To analyze the performance of our proposed gateway selection scheme, we simulate our scheme in NS2.30 [16] and the simulation parameters are shown in Table I. We study the topology of 25 nodes with 5 gateways over a 1000mu1000m area. Each node has a maximum transmission range of 175 meters. The network is simulated for 1000s. We assume that all gateway nodes are equipped with 802.11b interface and are accessed by MANET nodes that are within transmission range of each other or MANET nodes that have multi-hop path to gateway. The speed of MANET and gateway nodes is uniformly distributed between 0 and 1m/s, 2m/s, to 5m/s in a RandomWalk [15] fashion. The communication between MANET nodes and the infrastructure nodes is carried out in a Constant Bit Rate (CBR) manner. CBR packet size of 512 bytes and the CBR interval of 0.3, 0.15, to 0.009s are considered in the simulation. Average number of 5 CBR connections per minute with average connection duration of 80s is implemented in the simulation. The simulation results are the average of 3 randomly generated traffic models executed over each of the 5 randomly generated movement scenarios. The performance of the proposed gateway selection algorithm is compared with the gateway selection schemes proposed in [10] and [11], which are referred as conventional 1 and conventional 2, respectively, in this paper. Figure 5 to 10 show the simulated performances of the proposed and the conventional schemes.
TABLE II.
AVERAGE PERCENT PERFORMANCE IMPROVED BY PROPOSED SCHEME
Avg. Improvement Compared to Conv. 1, Į=0.05
Success % 63.8
64.2
Drop % 67.5
Conv. 1, Į=0.03
53.3
53.3
31.8
Conv. 2, Į= 0.05
5.0
5.1
8.3
Conv. 2, Į= 0.03
3.9
4.0
3.4
Conv. 1, 1m/s
36.9
30.1
51.8
Conv. 1, 3m/s
94.5
78.5
64.0
Conv. 2, 1m/s
3.7
6.6
6.8
Conv. 2, 3m/s
1.3
8.7
1.7
Average throughput and packet drop ratio are shown in Fig. 6 and Fig. 7, respectively. It shows that the proposed scheme improves the throughput and minimizes the packet drop ratio compared to the conventional schemes. Proposed scheme improves the throughput and reduces the drop ratio because, it considers normalized (values between 0 and 1) end-to-end QoS parameters i.e. path stability, maximum available capacity and path delay. It precisely computes the path stability by considering the epoch length of the individual nodes in the route, which reduces packet loss and improves the throughput. Table II summarizes the average percent improvement in the
Figures 5(a) and (b) show the CBR success ratio for varying node speed and CBR interval. It is evident from the graphs that the proposed scheme outperforms the conventional 1 and also has higher success ratio compared to conventional 2. The reason that conventional scheme 1 has less success ratio, because conventional 1 only selects a route with maximum
1
0.8
70
0.7
0.9
Throughput (Kbps)
0.4 0.3 0.2
50 40 30 20
0.1
10
0
0 1
2
3 Node Speed (m/s)
4
2
3 Node Speed (m/s)
(a)
0.8 0.7 0.6 0.5 0.4
110 90 80 70
1
2
3 Node Speed (m/s)
0.1
0 0.05
0.03
0.01
0.009
(b) Figure 5. Average Success Ratio versus (a) Node Speed and (b) CBR Interval [Į]
0.8
30 10
5
Conventional 1, 1m/s Conventional 1, 3m/s Conventional 2, 1m/s Conventional 2, 3m/s Proposed, 1m/s Proposed, 3m/s
0.9
40
0.2
4
(a) 1
50
20
0.1
0.5
5
60
0.3
0.15
4
Conventional 1, 1m/s Conventional 1, 3m/s Conventional 2, 1m/s Conventional 2, 3m/s Proposed, 1m/s Proposed, 3m/s
100
Throughput (Kbps)
Conventional 1, 1m/s Conventional 1, 3m/s Conventional 2, 1m/s Conventional 2, 3m/s Proposed, 1m/s Proposed, 3m/s
0.9
0.2
0.6
(a)
1
0.3
0.7
0.3
1
5
0.8
0.4
Avg. Packet Drop Ratio
Avg. Success Ratio
0.5
Avg. Packet Drop Ratio
60
0.6
Avg. Success Ratio
Throughput %
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
0.3
0.2
0.15
0.1
0.05
0.03
0.01
0.009
(b) Figure 6. Average Throughput versus (a) Node Speed and (b) CBR Interval [Į]
450
0.3
0.2
0.15
0.1
0.05
0.03
0.01
0.009
(b) Figure 7. Average Packet Drop Ratio versus (a) Node Speed and (b) CBR Interval [Į]
0.16
0.16
0.14
0.14 0.12 0.1 0.08 0.06 0.04
0.12
0.08 0.06 0.04 0.02
0
0
2
3 Node Speed (m/s)
4
5
Figure 8. Avg. End-to-End delay v/s node speed.
performance by proposed conventional schemes.
scheme
0.3
0.2
0.15
0.1
compared
to
the
The average energy consumption is shown in Fig. 10. It is evident that conventional 1 has less average energy consumption per node because due to unstable route most of the traffic is not carried out by this scheme. In result it has less energy consumption per nodes. The proposed scheme selects more stable and less congested routes compared to conventional 2 that consequently improves the throughput and also reduces control energy consumption and control overhead.
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
CONCLUSION
We have proposed a gateway selection scheme that considers multiple end-to-end QoS parameters such as: precise route availability period, route capacity and latency. The simulation results show that the proposed scheme improves the network throughput, success rate and reduces the packet drop ratio, end-to-end delay and energy consumption per node compared to the conventional schemes. ACKNOWLEDGMENT
[11]
[12]
[13]
This work is partly supported by the Grant in Global Center of Excellent Program “High-Level Program Cooperation for Leading-Edge Platform on Access Spaces” of Keio University from the MEXT, Japan. REFERENCES [1]
0.05
0.03
0.01
0.009
Figure 9. Avg. End-to-End delay v/s CBR interval.
Figures 8 and 9 depict the average end-to-end delay experienced by the data packets for different node speed and CBR interval, respectively. It shows that the proposed scheme has less end-to-end delay compared to conventional 2 and slightly higher end-to-end delay in contrast to the conventional 1 scheme. The reason of improved end-to-end delay performance by the proposed scheme compared to conventional 2 is that it considers end-to-end delay as one of the gateway selection parameter. Conversely, our scheme has slightly higher latency compared to conventional 1 because, conventional 1 forwards traffic through shortest but unstable, which results in a very high packet loss and reduced throughput.
IV.
140
0.1
0.02 1
150
Conventional 1, 1m/s Conventional 1, 3m/s Conventional 2, 1m/s Conventional 2, 3m/s Proposed, 1m/s Proposed, 3m/s
Avg. Energy Consumption / Node (j)
0.18
Avg. End-to-End Delay (s)
Avg. End-to-End Delay (s)
0.2 0.18
[14]
[15]
G. Aggelou, Mobile Ad Hoc Networks: From Wireless LANs to 4G Networks, McGraw-Hill Professional, November, 2004.
451
[16]
130 120 110 100 90 80 70 1
2
3 Node Speed (m/s)
4
5
Figure 10. Avg. energy consumption/node v/s speed.
E. Baccelli, K. Mase, S. Ruffino, and S. Singh, “Address autoconfiguration for MANET: terminology and problem statement,” Internet Draft, draft-baccelli-autoconf-statement-03. H. Luo, X. Meng, R. Ramjee, P. Sinha, and L. Li, “The design and evaluation of unified cellular and ad-hoc networks,” IEEE Trans. Mobile Computing, vol. 6, pp. 1060-1074, Sept. 2007. P. M. Ruiz, et al., "Internet connectivity for mobile ad hoc networks: solutions and challenges," IEEE Communication Magazine, pp 118-125, Oct. 2005. B. Shen, B. Shi, B. Li, Z. Hu, and L. Zou, “Adaptive gateway discovery scheme for connecting mobile ad hoc networks to internet,” in Proc. Wireless Communications, Networking and Mobile Computing, vol. 2, pp. 795-799, Sept. 2005. M.C. Domingo, “Integration of ad hoc networks with fixed networks using an adaptive gateway discovery protocol,” in Proc. 2nd IET International Conference on Intelligent Environments, July 2006. M. Ghassemian, P. Hofmann, V. Friderikos, C. Prehofer, and A.H. Aghvami, "An optimised gateway selection mechanism for wireless ad hoc networks connected to the internet," in Proc. IEEE 63rd Vehicular Technology Conference, May 2006. C. Shin, S.H. Kim, and S. An, "Stable gateway selection scheme based on MANET with internet," in Proc. Sixth IEEE International Conference on Computer and Information Technology, Sept. 2006. H. Ammari and H. El-Rewini, "Using hybrid selection schemes to support QoS when providing multihop wireless Internet access to mobile ad hoc networks," in Proc. First International Conference on Quality of Service in Heterogeneous Wired/Wireless Networks, Oct. 2004. B.N. Park, W. Lee, C. Lee, and C.K. Shin, "QoS-aware adaptive internet gateway selection in ad hoc wireless internet access networks," in Proc. 3rd International Conference on Broadband Communications, Networks and Systems, Oct. 2006. M. Wu, W.K.G. Seah, and L.W.C. Wong, “A link-connectivityprediction-based location-aided routing protocol for hybrid wiredwireless networks,” in Proc. 2nd International Conference on Mobile Computing and Ubiquitous Networking, April 2005. F.P. Setiawan, S.H. Bouk and I. Sasase, “An optimum multiple metrics gateway selection mechanism in MANET and infrastructured networks integration,” IEICE Transactions on Communications, Vol.E92-B, No.08, Aug. 2009. M. Ghassemian, P. Hofmann, C. Prehofer, V. Friderikos, and H. Aghvami, "Performance analysis of Internet gateway discovery protocols in ad hoc networks," in Proc. Wireless Communications and Networking Conference, March 2004. P. Ratanchandani and R. Kravets, “A hybrid approach to internet connectivity for mobile ad hoc networks,” in Proc.IEEE WCNC 2003, Vol. 3, pp. 1522–1527, March 2003. T. Camp, J. Boleng, and V. Davies, “A survey of mobility models for ad hoc network research,” in Proc. Wireless Communication & Mobile Computing (WCMC): Special issue on Mobile Ad Hoc Networking: Research, Trends and Applications, vol. 2, no. 5, pp. 483-502, 2002. “The Network Simulator – NS-2”, http://www.isi.edu/nsnam/ns/
