Research Motivation - 應用於平均負載之移動式隨意網路繞徑協定

CHAPTER 1 Introduction

2.3 Research Motivation

Although several ad hoc routing protocols have been proposed, these protocols only consider traffic load of node but do not precisely calculate total traffic load of shared radio channels within specified collision domains. In DLAR the load metric of a node is defined as the number of packets buffered in the node, and the load metric of a route is the summation of the load metrics of the nodes on that route. The DLAR protocol does not optimally reflect the actual load since buffered packets may vary in size. Besides, this protocol does not consider the effect of contention/collision within shared channels. In LBAR, the load metric of a node is the total number of routes flowing through the node and its neighbors. Although LBAR takes traffic loads of contending neighbors into account, it is not optimal since it does not account for the various traffic sizes of each route.

In this research, we aim to propose a load-balanced ad hoc routing protocol, whose load metric of multi-hop route considers the effect of contention/collision within shared channels. A novel load estimating method is defined. Based on the proposed method for load estimation, the proposed routing protocol explores the least load routes, which provide high packet delivery rate and sufficient decrease the packet latency for all source and destination pairs in MANETs.

CHAPTER 3 The Proposed Load-Balanced Routing Protocol

The proposed load-balanced routing protocol is named as Ad-hoc On-demand Load-Balancing (AOLB) routing protocol. This protocol aims to increase packet delivery rate and decrease packet delay by discovering the least loading route using a novel load estimation method and a load-aware path finding algorithm. In Fig.3-1, the flow diagram of the AOLB routing protocol is depicted. Three major processes form the proposed protocol, which are Load Estimation, Route Discovery, and

Route Recovery. The following paragraphs are given to describe the

details of these processes.

Fig. 3-1 Flow chart of the proposed AOLB routing protocol,

3.1 Load estimation process

In this process, each node of MANETs begins to estimate the link loads of radio links to its neighbors. To reflect the effect of contention in shared channels, here the link load of a given radio link i at time t, Li(t), is equal to the ratio of averaged traffic size pass through the collision domain of radio link i to the maximum data rate of this link. Assuming the symmetric MAC protocol is used in MANETs, therefore, the link load is estimated by

i adjacent nodes connected by link i; Tp(t) and Tq(t) are summation of averaged traffic size pass through node p, node q and their neighboring nodes within a time window δ, which is represented by

S

(t-kΔt) represents the total traffic size pass through node m between

time t-(k+1)Δt and time t-kΔt and is defined by

)

transmitted by node m between time p-Δt and time p, respectively;

δ

is the total number of unit time steps of time window to calculate the averaged traffic size;

Δt represents a unit time step length;

N

_m is a node set formed by all neighboring nodes of node m;

In order to keep track of the link load between a given node and its neighbors, each node periodically broadcasts Hello message (HELLO) attached with traffic size passed through itself, the Local Load field calculated from Eq.(3), and the summation of traffic sizes passed through

its neighbors and itself, the Regional Load field calculated from Eq.(2).

Fig. 3-2 depicted the packet format of proposed Hello message. With the received HELLOs, the receiving node is aware of traffic loads nearby and determines link loads of adjacent radio links. The detail procedure of load estimation process is presented by following steps:

Fig 3-2 Packet format of proposed Hello message used by AOLB

Step 1: Counting traffic size of every node at each unit time. The

averaged traffic size passed through each node at recently unit time, including the received traffic and the transmitted traffic, is temporarily counted by Eq.(3) and recorded. Then, the averaged traffic size of that node would be filled in the Local Load field of the broadcasting HELLO message.

Step 2: Determining total traffic size of every node and its neighbors.

With the received HELLO messages, averaged traffic sizes of all

neighboring nodes together with their IP addresses and timestamps are recorded in a table, named as Regional Loading Status Table (RLST) at each receiving nodes. With the memorized traffic sizes of neighboring nodes in RLST, the receiving node further evaluates the total traffic size of its neighbors and itself by Eq.(2) with predefined time window δ. The summation of traffic sizes is filled in the Regional Load field of HELLO messages and broadcasted to its neighbors per unit time.

Step 3: Estimating and recording the link loads. Due to the summation of

averaged traffic sizes of a specified node p, says Tp(t), and that of its neighboring node q, says T_q(t), are determined, the link load of radio link between node p and q is estimated by Eq.(1). Therefore, the link loads of all adjacent radio links are estimated and recorded in a table, named as

Neighbor Link Load Table (NLLT), which is used to keep track the

up-to-date link loads of radio links to its neighbors for both route discovery and route recovery processes. The link loads are estimated per unit time and the NLLT are periodically updated.

3.2 Route Discovery Method

This process executes a path-finding algorithm to discover the multi-hop route with minimum route load for a specified source and destination pair. The following steps present the detail procedures. In steps 1-3, multiple available routes are explored at the destination node.

The load of each available route is also determined within this stage. In

node. With the intention of transmitting packets at source node, the source node, intermediate nodes, and the destination node perform the following steps:

Step 1: Source node starts to find out the least loaded route by broadcasting the RREQ message. To fulfill dynamically occurred route

request for a specified source and destination pair, the source node initiates a route discovery process by generating a Route Request (RREQ) message and broadcasting it to all its neighboring nodes. Fig. 3-3 depicts the packet format of proposed RREQ message, which is similar to the RREQ in AODV routing protocol but includes a newly field named Route Load (which initially set to 0 at source node).

Fig 3-3 Packet format of proposed RREQ message used by AOLB

Step 2: Intermediate nodes restrict the flooding of RREQ message to

reduce control overhead. An intermediate mobile node might receive

multiple RREQs and rebroadcast these messages during the route discovery process. To decrease the control overhead caused by the flooding RREQ messages in MANETs, the intermediate mobile nodes

rebroadcast only the RREQ with the minimum value in Route Load field among the received RREQ messages.

Step 3: Nodes calculate route load of eligible routes from the source to the intermediate node. Before the intermediate node rebroadcasts the

selected RREQ message, it needs to recalculate the Route Load. Here the route load of a specified multi-hop route r at time t is given by

]

where i denotes the link within the route r and P_r represents the link set containing all links of the route r. As we mentioned in load estimation process, the link loads are collected from the NLLT of intermediate nodes along this route.

Step 4: Destination node select the route with the minimum Route Load.

With multiple RREQs continually arriving at destination node, the destination node only receives the RREQs within a time window, which starts from the first arrival RREQ. After the destination node determines the one with the minimum route load, it creates a Route Reply (RREP) message formatted similarly in AODV protocol for responding with the RREQ. Then the RREP is forwarded to the neighbor from whom the selected RREQ was sent.

Step 5: Intermediate nodes react RREP by building up the forwarding

entries in the route tables. When an intermediate node receives the RREP

message, it sets up the forward entry to the destination node in its route table. After processing the RREP message, the intermediate node forwards the RREP message toward the source node along the reverse route through which the selected RREQ message passed.

Step 6: Source node discovers the selected route and start to transmit data. The source node receives the RREP and sets up the route table to

the destination node. The least loaded route with minimum route load is discovered by a specified RREQ and the forward entries of route tables from source node through intermediate nodes to the destination are set up by the replying RREP.

3.3 Route Recovery

Due to the time-varying topologies of MANETs, the availability of the connecting routes would change with time. In order to react the change of route availability, a Route Recovery process is proposed.

Step 1: Using the periodically hello message to detect the load condition

AOLB routing protocol uses the Hello message to monitor link load to all neighbors. All mobile nodes periodically send Hello messages which containing the load information to its neighbors. Whenever a node receives a broadcast from a neighbor, it calculates interference load which is from its neighbor nodes. The interference load will periodically updates to react the change of load condition.

Step 2: Route error message, and route rediscover

If a link breakage is detected, a Route Error message is send to the

source node from the mobile node that detects the link breakage occurrence. When the source node receives the RERR message it will rediscover a new least loaded route by flooding newly issued RREQs to the destination node.

CHAPTER 4 Simulation Results and Analysis

In order to evaluate the performance of proposed AOLB routing protocol, the NCTUns network simulator [22] is employed. For comparing between the proposed AOLB protocol and related works, the AODV, DLAR and LBAR routing protocols are also included. As to the physical layer and MAC layer protocols, the IEEE 802.11b PHY and IEEE 802.11 DCF MAC protocol are selected. All of chosen protocols are evaluated on a simulated free-space propagation environment with a large-scale median path-loss model and a small-scale fading model . A traffic generator, which transmits UDP packets with specified constant packet delivery rate, is used to obtain the communication performance between the specified S-D pair. The random waypoint mobility model is chosen to emulate the moving behavior of each mobile node. Here, two significant metrics are considered to determine the performance of load-balanced MANET routing protocols in our simulations.

z Averaged packet delay (APD): the averaged packet delay is the averaged end-to-end delay time of all received data packets that are transmitted from the source nodes to the destination nodes.

Only end-to-end packet delays of successfully delivered packets are considered in our simulations.

z Packet delivery ratio (PDR): the packet delivery ratio P is the ratio of total number R of received data packets at destination nodes to the total number S of data packets transmitted by source

nodes during a specified time period, that is, P=R/S. In most of the simulation cases, the number R is less than the number S due to packet loss along the wireless multi-hop route.

In this paper, effects of time-varying topologies, traffic loads, and node density on APDs and PDRs are investigated and simulated. In most of simulation cases, 50 nodes are randomly distributed in a restricted 1500m × 300m rectangular area at the beginning. For each simulation case, total of 5 tests were made, where each test lasts for 500 seconds and samples the number of received packets and their packet delays per second in order to calculate both APD and PDR. The configurations of simulation parameters for most of simulation cases are depicted in Table 4-1.

Table 4-1. General configuration of simulation parameters

Parameters Parameter assigned values

Area of operation (m²) 1500 × 300

Number of nodes 50

Initial topology Randomly assigned

Moving speed of nodes (m/s) 10

Optimal radio channel capacity (Mbps) 2 Variance of small-scale fading effect (db) 20 Transmission range of single radio link (m) 200 Received power sensitivity and threshold (db) -74

Simulation time (sec) 500

4.1 Time-varying topology effect

In this simulation, we investigate the performance of load-balanced

routing protocols with different degree of node mobility as well as time-varying topology of MANETs. Total of 30 traffic flows are simultaneously generated with 5 packets/second packet transmission rate.

These traffic flows are further divided into four groups by the size of transmitted packet. They are 8 flows with 128 bytes per packet, 8 flows with 256 bytes per packet, 7 flows with 512 bytes per packet, and 7 flows with 1024 bytes per packet. In order to grade the degree of node mobility, the random waypoint mobility model with configurable waypoint pause

time [21] is chosen to emulate the moving behavior of mobile nodes. The

waypoint pause time here indicates the waiting time period between the time while the mobile node arriving a specified waypoint, and the time while it starts to leave the waypoint. The smaller the pause time, the higher degree of node mobility would be. Fig. 4-1 and Fig. 4-2 show the APD and PDR versus node pause time by using selected routing protocols, respectively.

Fig 4-1 Averaged packet delay vs. waypoint pause time by using AODV, AOLB, DLAR, and LBAR.

0.5

Fig 4-2 Packet delivery ratio vs. waypoint pause time by using AODV, AOLB, DLAR, and LBAR.

It is found that the proposed AOLB protocol always performs shortest APD in all test cases and provides highest PDR when the node pause time is larger than 300 seconds. From Fig. 4-1, there is a trend that the APD decrement of AOLB to other protocols increases with the increasing of waypoint pause time. The largest APD decrements of AOLB are 4.14 seconds, 0.88 seconds, and 1.28 seconds to AODV, DLAR, and LBAR protocols, respectively, when the pause time is 500 seconds. Fig.

4-2 shows that the PDR improvement of AOLB over other protocols increases at the same time. The best PDR improvement of AOLB are 12.59%, 4.38%, and 6.51% over AODV, DLAR, and LBAR protocols, respectively. This phenomenon indicates that the AOLB protocol provides better performance than other protocols when the degree of node mobility is low. Due to in highly variant network topologies, it is hard to lower down the probability of route breakage and difficult to balance traffic loads in nature, so that either the AOLB nor other load-balanced routing

shortest-path based AODV protocol.

4.2 Traffic load effects

In this simulation, we investigate the traffic load effects on the performance of load-balanced routing protocols. Two simulation scenarios are considered. In the first scenario, the traffic load effect caused by the increasing packet transmission rate on each traffic flow is investigated. Total of 30 traffic flows are simultaneously generated with a specified packet transmission rate. These traffic flows are further divided into four groups by the size of transmitted packet similar to prior simulations presented in section 4.1. The waypoint pause time of each moving path of nodes is 250 seconds. To vary traffic load of each flow, packet transmission rates are increased from 1 packets/second to 8 packets/second. Fig. 4-3 and Fig. 4-4 show APD and PDR versus packet transmission rate by using AODV, DLAR, LBAR, and AOLB protocols, respectively.

Fig 4-3 Averaged packet delay vs. packet transmission ratio by using AODV, AOLB, DLAR, and LBAR.

Fig 4-4 Packet delivery ratio vs. packet transmission rate by using AODV, AOLB, DLAR, and LBAR.

From Fig. 4-3, it is found that the averaged packet delay increases with packet transmission rate by whatever the selected routing protocol.

This phenomenon is caused by the increasing packet contention time at each intermediate node due to the increment of traffic loads. By

comparing all simulated protocols, the proposed AOLB protocol always selects the routes pass through minimum loaded collision domains, which effectively decreases packet contention time as well as APD. The APD using proposed AOLB protocol is better than those of AODV, DLAR, and LBAR for all transmission rates. The largest APD decrements of AOLB are 4.01 seconds, 1.44 seconds, and 2.19 seconds to AODV, DLAR, and LBAR protocols, respectively, when the transmission rate is 8 packets/second. It is noted that the AODV protocol, which is based on shortest path algorithm, cannot evenly distribute traffic loads as a result producing traffic congestion and performing highest APD in all test cases.

Fig. 4-4 shows that the packet delivery ratio decreases with the increasing of packet transmission rate. The AOLB protocol provides best PDR when the packet transmission rate is larger than 3 packets/second. The average improvements of PDR are 11.9%, 3.12%, and 6.24% over AODV, DLAR, and LBAR, respectively.

In the second scenario, the traffic load effect caused by the increasing number of traffic flows is investigated. For a specified number of traffic flows, all traffic flows are evenly divided into four groups by the size of transmitted packet, including 128 bytes per packet, 256 bytes per packet, 512 bytes per packet, and 1024 bytes per packet. These traffic flows are simultaneously generated with 5 packets/second packet transmission rate.

Four numbers of traffic flows are considered, including 10 flows, 20 flows, 30 flows, and 40 flows. Fig. 4-5 and Fig. 4-6 show APD and PDR versus number of traffic flows by using selected routing protocols, respectively.

Fig 4-5 Averaged packet delay vs. number of traffic flows by using AODV, AOLB, DLAR, and LBAR.

Fig 4-6 Packet delivery ratio vs. number of traffic flows by using AODV, AOLB, DLAR, and LBAR.

From Fig. 4-5, it is found that the averaged packet delay increases with number of traffic flows. This phenomenon is also caused by the increasing of packet contention time when the whole traffic load in the

AOLB protocol has minimum APD in all test cases. Although AOLB perform much better APD than AODV, however, the AOLB cannot significantly improve other related works, DLAR and LBAR, when the number of traffic flows is quite small (10 flows) or large (40 flows). The largest APD decrements of AOLB are 3.54 seconds, 0.39 seconds, and 0.71 seconds to AODV, DLAR, and LBAR protocols, respectively, when the number of traffic flows is 30 flows. Due to the major difference between AOLB and these related works is the estimation method of load metric for selecting multi-hop route, this phenomenon indicates that estimation method of load metric would not be the critical factor for load-balanced routing protocol in lightly-loaded or fully-loaded MANETs.

Fig. 4-6 shows that the packet delivery ratio decreases with the increasing of packet transmission rate. The average improvements of PDR are 11.4%, 3.48%, and 7.45% over AODV, DLAR, and LBAR, respectively.

4.3 Node density effect

Here, to investigate network performance with different node density, 50 mobile nodes are configured to randomly move within a 600m^×300m, 900m ^× 300m, 1200 ^× 300m, 1500m ^× 300m, 1800m ^× 300m, and 2100m^×300m area. Total of 30 traffic flows are simultaneously generated with 5 packets/second packet transmission rate. Fig. 4-7 and Fig. 4-8 show the APD and PDR versus node density by using selected routing protocols, respectively.

node density (per myriare)

Average Packet Delay(sec)

AODV AOLB DLAR LBAR

Fig 4-7 Packet delivery ratio vs. node density by using AODV, AOLB, DLAR, and LBAR.

Fig 4-8 Packet delivery ratio vs. node density of traffic flows by using AODV, AOLB, DLAR, and LBAR.

Fig. 4-7 shows that the largest APD decrements of AOLB are 3.54 seconds, 0.39 seconds, and 0.71 seconds to AODV, DLAR, and LBAR protocols, respectively. From Fig. 4-8, AOLB provides the best PDR

comparing to other protocol. The averaged improvement ratios of PDR using AOLB to the PDR using AODV, DLAR, and LBAR are 10.72%, 2.24%, 6.22%, respectively. There is a trend that the PDR decreases with the decreasing of node density. This is caused by the averaged hop count of routes is large when node density is low, which results in high probability of packet loss and lowers down the PDR.

CHAPTER 5 Conclusions

In this paper, a load-balanced ad hoc routing protocol, the Ad-hoc On-demand Load-Balancing (AOLB) routing protocol, for MANETs is proposed. The protocol is based on a novel link/route load metric to evenly distribute traffic loads in MANETs, which aims to avoid packet contention/collision in collision domains. In AOLB protocol, every node in an MANET maintains both a routing table and a data sheet with estimated traffic load of its belonging collision domain, which slightly increases the computational complexity. Simulation results by changing the degree of node mobility, the total traffic load within a network, and node density indicate that the AOLB leads to significant averaged packet delay decreases about 1 to 4 seconds over AODV and averaged packet delivery ratio increases about 7% to 14% over AODV. The improvement of AOLB also dominates over that of DLAR and LBAR. It is found that

在文檔中應用於平均負載之移動式隨意網路繞徑協定 (頁 28-0)