Вы находитесь на странице: 1из 6

Sains Malaysiana 38(3)(2009): 423428

Application of Taguchis Design of Experiment in Performance Analysis of Destination Sequence Distance Vector (DSDV) Routing Protocol in Mobile Ad Hoc Networks
(Penggunaan Rekabentuk Ujikaji Taguchi bagi Analisis Prestasi Laluan Protokol DSDV dalam Rangkaian Mudah Alih Ad Hoc) HAZURA MOHAMED*, MUHAMMAD HISYAM LEE, MAZALAN SARAHINTU, SHAHARUDDIN SALLEH & BAHROM SANUGI

Numerous routing protocols for mobile ad hoc networks (MANETs) have been designed in process information delivery from a source node to a destination node. In this paper, the Taguchis design of experiment (TDE) has been applied to investigate the performance of Destination Sequence Distance Vector (DSDV) routing protocol in MANETs. The effects of network parameters namely terrain sizes, node speeds, network sizes, transmission ranges, transmission rates, pause times and the number of maximum connections on packet delivery ratio and routing overhead in medium scale ad hoc networks have been done through simulation experiments. Through this study, we can rank these factors that may affect packet delivery ratio and routing overhead. The response performance was analyzed based on signal-to-noise ratio and analysis of variance (ANOVA). The results revealed that the transmission range was the most influential factor on the packet delivery ratio, followed by terrain size and transmission rate. The network size had the greatest effect on routing overhead, followed by the transmission range. Keywords: Mobile ad hoc networks; routing protocol; Taguchis design of experiment Laluan protokol bagi rangkaian mudah alih sementara (RMS) direka bentuk untuk proses penghantaran maklumat daripada nod sumber kepada nod destinasi. Dalam kajian ini, reka bentuk ujikaji Taguchi telah digunakan untuk menyelidik prestasi protokol DSDV dalam RMS. Pengaruh saiz terain, kelajuan nod, saiz rangkaian, julat transmisi, kadar transmisi, masa berehat dan bilangan hubungan maksimum ke atas nisbah penghantaran padatan dan overhed laluan dalam senario rangkaian mudahalih bersaiz sederhana telah dibuat secara ujikaji bersimulasi. Melalui kajian ini, susunan faktor mengikut keutamaan masing-masing, berdasarkan pengaruhnya terhadap dapatan nisbah penghantaran padatan dan overhed laluan dapat dikenal pasti. Data yang diperoleh daripada ujikaji bersimulasi telah dianalisis berdasarkan nisbah isyarat-kepada-kebisingan dan analisis varians (ANOVA). Hasil kajian menunjukkan faktor yang paling mempengaruhi nisbah penghantaran padatan ialah julat transmisi, diikuti saiz terain dan kadar transmisi. Manakala saiz rangkaian merupakan faktor yang sangat mempengaruhi overhed laluan diikuti saiz transmisi. Kata kunci: Laluan protokol; rangkaian mudah alih sementara; reka bentuk ujikaji Taguchi INTRODUCTION A mobile ad hoc network (MANET) is a dynamic wireless network that can be formed with the absence of infrastructure such as base stations or access points. Communication between two nodes is only possible when they are within their radio communication range otherwise they need router to forward packet from source to destination nodes. In such network a node can act as a host and a router in the process of transmission information. Some examples of the possible uses of MANET include in military application, conferences participants in sharing information using notebook and in interactive lecture where students use laptops. Designing robust routing protocols which must have the ability to adapt the frequently and randomly changing network topology is the special challenges in the dynamic nature of MANET. A number of routing protocols have been proposed and a review of a range of protocols in such a network is given in (Abolhasan et al. 2004). Several performance evaluations of MANET routing protocol have been performed over the past few years. Most of the previous studies evaluate the performance of routing protocols used simulation based (Ari et al. 2000; Boukerche 2004; Das et al. 2000; Lakshmi & Sankaranarayan 2005). 75.5% of MANET research published in the 2000-2005 proceedings of the ACM International Symposium on Mobile Ad Hoc
ABSTRAK

ABSTRACT

424

Networking and Computing (MobiHoc) used simulation to test their research (Kurkowski 2005) and many of them have done through trial-and-error simulation experiment. Their results are also lacking of statistical analysis even though they have some reasonable results. In this paper, we carry out a systematic procedure to measure the ability of Destination Sequence Distance Vector (DSDV) routing protocol to react network topology changes in terms of terrain size, mobile nodes movement, and network size, number of connections among nodes, transmission range and also the transmission rate among nodes. Using Taguchis design of experiment (TDE) we quantify the effect of these factors on DSDV protocols performance or in other words, whether the DSDV routing protocol is able to route and deliver each data packet to the destination node. The use of TDE to analyze and estimate the effects of MANETs routing protocols performance has been initially created by Lee (2003). DSDV (Perkins & Bhagwat 1994) is a one of tabledriven routing protocols, which is an attempt to maintain optimal routes to all destinations at all times, regardless of whether they are needed. To support this, the routing protocol propagates information updates about a networks topology throughout the network. In DSDV each node maintains a routing table that list all available destination and routes to them. The sequence number is used to distinguish stale routes from new ones and thus avoid the formation of loops. When DSDV detects a broken route, it stores the packet until the next route update at which time it tries to resend the packet. MATERIALS AND METHODS
TAGUCHIS DESIGN OF EXPERIMENT

we choose two quality characteristics as indicators of the DSDVs performance, namely packet delivered (PDR), and routing overhead (RO). The data packet delivery ratio is the ratio of the number of data packets delivered to the destination nodes divided by the number of data packets transmitted by the source nodes. Routing overhead is the ratio of forwarded (hopped) packets to sent ones. Greater PDR and smaller RO are indication of better performance. Therefore, for obtaining the best performance of DSDV, the GB and SB ratios were selected for PDR and RO, respectively. The SNR for PDR and RO can be calculated as follows (Peace 1992): (1)

(2) where r is the number of simulation repetitions under the same experimental number and y is the response performance. Seven factors, namely terrain, node speed, network size, transmission range, transmission rates, pause time and maximum connection were considered for the experiment. Each factor has two levels, denoted as low (1) and high (2). Sarahintu et al. (2007) analyzed the effect of these factors on the performance of different protocol, Dynamic Source Routing (DSR) in small area network, whereas, in this work we focus on medium area network. Table 1 summarized the factors and their levels. To select an appropriate orthogonal array for the simulation experiments, the total number of degrees of freedom (DF) need to be computed. The degrees of freedom are defined as the number of comparisons that needs to be made to determine which level is better. Since each two-level factor has 1 degree of freedom, the DF required for seven factors, each at two levels is 7. Taguchis orthogonal array (OA) are selected on the basis of the condition that the total DF of the selected OA should be greater than or equal to the total DF required for the experiment (Ross 1996). Hence a L8 orthogonal array was adopted in this study as presented in Table 3. We have 8 rows which represent the number of experiments
TABLE

Taguchis design of experiment (TDE) approach offers a way for the researcher to determine, before the runs are made, which specific configurations to simulate so that the desired information can be obtained with the least amount of simulation (Law & Kelton 2000). It is very effective to deal with response influenced by many factors. The difference between a traditional full factorial design of experiment and TDE is the significant reduction in size of experiment. Taguchi method uses signal-to-noise ratio (SNR), which is denoted as , instead of the average value to interpret the experimental results data into a value for the evaluation characteristic in the optimum setting analysis (Ross 1996). This is because the SNR can reflect both average and variation of the quality characteristics. Depending on the quality characteristics involved, different SNRs may be applicable, including smaller is better (SB), nominal is best (NB), or greater is better (GB). Regardless of the category of the performance characteristic, the greater SNR corresponds to the better performance characteristics. Therefore, the best level of the factors is the level with the highest SNR. The SNR is expressed in decibels (dB) and treated as response of the experiments. For this study,

1. Selected factors and assigned levels 1 Level 2

Label A B C D E F G

Factors

Terrain size (m2) 200 200 Node speed (ms-1) 0.72 Network size 50 Transmission range (m) 30 Transmission rates (packets/s) 1 Pause time (s) 60 Maximum connection 2

100 400 1.34 100 100 10

200 8

425
TABLE

2. The L8 orthogonal array B 1 1 2 2 1 1 2 2 C 1 1 2 2 2 2 1 1 Factors D 1 2 1 2 1 2 1 2 E 1 2 1 2 2 1 2 1 F 1 2 2 1 1 2 2 1 G 1 2 2 1 2 1 1 2

Experiment No. 1 2 3 4 5 6 7 8

A 1 1 1 1 2 2 2 2

response in analyzing the performance of the DSDV routing protocol. In order to obtain results that can give valuable information, we replicate the eight simulations five times, this results in a total of 40 simulation runs. Each simulation experiment was executed for 900 seconds. Then SNR for PDR and RO is computed using equation (1) and (2), respectively for each experiment as shown in Table 3. The overall mean for SNR of PDR is found to be 34.243 db whereas overall mean for SNR of RO is obtained as -83.778 db.
RESULTS AND DISCUSSIONS LEVEL AVERAGE RESPONSE ANALYSIS

that we will conduct, and 7 columns indicating the factors involved. For every factor, four tests would be carried out at one level, while the remaining four would be performed at the other level. By using Taguchi experimental design we can simultaneously evaluate the effects of seven factors with only 8 experiments rather than 128 experiments in a traditional full factorial experimental design.
SIMULATION ENVIRONMENT

The level average response analysis is done by averaging the SNR value at each level of each factor and plotting the values in a graphical form. The average SNR, which is denoted as of each level are calculated using formula (3) and are shown in Table 4 and 5, respectively. (3) where i is the SNR for the ith experiment involving factor F at level L. For example, the computation for average SNR for PDR of the terrain at low level (1) is shown below:

In this work, simulation sessions have been performed using the network simulator software (NS-2.29) for wireless networks. We use NS-2 for our simulations because it has been used in various works in the literature and is due to the fact that DSDV is already implemented in the simulator. NS-2 is a discrete event simulator targeted at networking research and written in C++, with an OTcl/ Tk interpreter as a front-end. The simulation of the mobile network has been carried out on an Intel Core 2 Duo CPU processor at 2.33 GHz, 2046 MB of RAM and running Linux Fedora Core 3. The objective of our simulations is to measure the ability of DSDV routing protocol to react to the frequent and random changes of network topology. To achieve this objective, we quantify the effects of those seven factors aforementioned with respect to PDR and RO as a measure
TABLE

Higher average SNR between the two levels of each factor implies the optimal level of the factor. As shown in Table 4, the optimal level for factor A is at low level (1), A1. Besides, high level (2) is optimal level for factor B, B2. The absolute difference between the average SNR values of the two levels reveals the effect of each factor. The larger the difference, the stronger is the influence to the performance. The average SNR for greater the better for packet delivery ratio and average SNR for smaller the better for routing overhead are shown in Table 4 and Table 5, respectively. To be clear, a main effect plots can be

3. SNR for PDR and RO Performance metrics


(SNR)

Experiment no. A 1 2 3 4 5 6 7 8 Total 1 1 1 1 2 2 2 2 B 1 1 2 2 1 1 2 2

Levels of factors C 1 1 2 2 2 2 1 1 D 1 2 1 2 1 2 1 2 E 1 2 1 2 2 1 2 1 F 1 2 2 1 1 2 2 1 G 1 2 2 1 2 1 1 2

PDR 30.71 39.83 36.48 39.68 25.41 39.69 22.67 39.47 273.94

RO -80.76 -74.32 -95.17 -81.88 -95.67 -85.85 -79.38 -77.19 -670.22

426

used to visualize performance changes as each individual factor level is changed. Figure 1 and Figure 2 show the response graph for SNR of packet delivery ratio and routing overhead. The slope of the line shift indicates the degree of particular factor effect on the performance metrics. The greater the slope value, the greater the magnitude of the main effects. Study of Table 4 and Figure 1 suggests that transmission range (factor D), terrain size (factor A) and transmission rates (factor E) can be treated as significant factors whereas node velocity (factor B), network size (factor C), pause time (factor F) and maximum connection (factor G) are insignificant. The lower level value for terrain size (A1) and transmission rate (E1) and higher level value for transmission range (D2) appears to be the best choice to get higher value of packet delivery ratio. Since node velocity (factor B), network size (factor C), pause time (factor F) and maximum connection (factor G) are insignificant, they could be set at any level. However, by looking at Table 4, the high level value for these factors produces higher SNR.
TABLE

insignificant factors which refer to the level that will give a higher SNR, thus will get the lower routing overhead.
TABLE

5. SNR of each factor level for RO Level 2 -84.525 -83.408 -89.644 -79.810 -82.813 -83.680 -85.589 Level 1-Level 2 Optimal level 1.491 0.743 11.730 7.938 1.932 0.199 3.620 1 2 1 2 2 2 1

Factor A B C D E F G

Level 1 -83.034 -84.151 -77.914 -87.748 -84.745 -83.879 -81.969

4. SNR of each factor level for PDR Level 2 31.810 34.575 35.315 39.668 31.898 34.668 35.298 Level 1-Level 2 4.865 0.665 2.145 10.850 4.690 0.850 2.110 Optimal level 1 2 2 2 1 2 2
FIGURE

Factor Level 1 A B C D E F G 36.675 33.910 33.170 28.816 36.586 33.818 33.188

2. Average SNR by factor level for RO

The purpose of the analysis of variance (ANOVA) is to investigate which factors significantly affect the performance characteristics. This is performed by separating the total variability of the response performance, which are measured by the sum of the squared deviations from the total mean of the response performance, into contributions by each of the factors under investigations and the error. The total sum of squares, SST from SNR can be calculated as: (4) where i is the SNR for the ith experiment. The total sum of squared deviation SST is partitioned into two sources, which are the sum of squared deviations due to each factor (SS), and the sum of the squared error. The SS can be calculated as: (5) where F represent one of the experiment factors, L is the level number of this factor F, e.g., L=2, t the repetition of each level of the factor F, e.g., t=4, and the sum of the SNR involving this factor F at level j. The present problem is a saturated design type of problem where all columns in OA are assigned with the

ANALYSIS OF VARIANCE (ANOVA)

FIGURE

1. Average SNR by factor level for PDR

Study of Table 5 and Figure 2 suggests that network size (factors C) and transmission range (factor D) have significant effect whereas terrain size (factor A), node velocity (factor B), transmission rates (factor E), pause time (factor F) and maximum connection (factor G) have least significant effect on lower of routing overhead. The low level value of factor C (C1) and the high level value of factor D (D2) appear to be the best choice to get the lower value of routing overhead. Similar to the packet delivery ratio, to find the best factor-level combinations for

427

factors; the variation due to error is estimated by pooling the factors having less influence for correct interpretation of results. By comparing the sum of squares values of each factor with the factor that has the highest sum of squares, which is less than 10% or lower than the most influential factor, then these factors are considered insignificant (Roy 2001). The sum of squares of error (SSe) is given by: (6) where nf is the number of experiment factors and SSj is the sum of squared deviations for each factor. The percentage contribution by each of the factor in the total sum of squares SST can be used to evaluate the importance of the factors on the performance characteristics. The values indicate the relative power of a factor to reduce variation. For a factor with a higher percent contribution, a small variation will have a great influence on the performance. The of each factor is given as (7) where is defined as the sum of squares of factors minus the error variance times the degree of freedom of each factor. The F-test can also be used to determine which
TABLE

factors have significant effect on the performance. Usually, the factors have a significant effect on the performance metrics if F value is large. Table 6 and 7 show the results of the ANOVA for the PDR and RO, respectively. From Table 6, the transmission range was found to be the major factor affecting PDR (66.35%), whereas the terrain size and transmission rates affect the PDR by 12.17% and 11.20%, respectively. At the 95% confidence level (F0.05,1,4 = 7.71), the transmission range, network size and transmission rates significantly effect on the performance of packet delivery ratio. Table 7 shows the results of the ANOVA for RO. It can be found that the network size is the most significant factor for affecting routing overhead (60.69%) followed by transmission range (26.82%). At the 95% confidence level (F0.05,1,5 = 6.61), the network size and transmission range significantly effect on the performance of routing overhead. CONCLUSION In this paper, a systematic procedure based on the Taguchis design of experiment (TDE) has been proposed to quantify the effect of terrain size, node speed, network size, transmission range, transmission rate, pause time and maximum connection on the performance of DSDV

6. ANOVA Table for PDR SS 47.349 0.882 9.183 235.418 43.985 1.445 8.865 {20.375} 347.127 MS 47.349 0.882 9.183 235.418 43.985 1.445 8.865 (5.094) F-ratio 9.296 46.217 8.635 12.173 66.352 11.204 10.272 100.00

Factor Terrain size (A) Node speed (B) Network size (C) Transmission range (D) Transmission rate (E) Pause time (F) Maximum connection (G) Error Pooled error Total

DF 1 {1} {1} 1 1 {1} {1} 0 {4} 7

TABLE 7. ANOVA

table for RO MS 4.446 1.105 275.200 126.011 7.465 0.080 26.205 7.860 F-ratio 35.013 15.032

Source Terrain size (A) Node speed (B) Network size (C) Transmission range (D) Transmission rate (E) Pause time (F) Maximum connection (G) Error Pooled error Total

DF {1} {1} 1 1 {1} {1} {1} 0 {5} 7

SS 4.446 1.105 275.200 126.011 7.465 0.080 26.205 {39.301} 440.512

60.689 26.821

12.490 100.00

428

routing protocol. Two quality characteristics were used as indicators of the DSDVs performance in our simulations. They are packet delivery ratio (PDR) and routing overhead (RO). The level of importance or the relative contributions of those factors on the PDR and RO is determined by using ANOVA. Based on ANOVA, a radio transmission range of 100 m was found to be significant for the best PDR and RO. A larger transmission range increases the distance progress of data packets toward their final destinations and the routing overhead decreases with the increase in the transmission range as a larger transmission range is accompanied with a smaller frequency of route discoveries. DSDV has better PDR with low transmission rate because packet collision is more severe when the packet size is large and square area size where mobile nodes can move. DSDV has smaller RO with small number of mobile nodes in the network. Since the selection of value level of factors is sensitive, it is important to take note that all the findings in this paper were totally based upon the factor levels considered in this study and may vary if different factor levels were used. In this work we executed several simulations in reducing potential variation when numerous factors are involved in the experiment.
REFERENCES

Wireless Networks. Asian Journal of Information Technology 4: 1195-1200. Law & Kelton 2000. Simulation, Modelling and Analysis. New York: McGraw-Hill Higher Education. Lee, M.H. 2003. Formulation of Properties for Motion Prediction in Wireless Networks and Simulation Analysis of Dynamic Destination-Sequenced Distance-Vector Protocol, Ph.D. Thesis, University of Technology Malaysia. Peace, G.S. 1992. Taguchi Method: A Hands-on Approach. Reading, Mass.: Addison-Wesley Publishing Company. Perkins, C. & Bhagwat, P. 1994. Highly Dynamic DestinationSequenced Distance-Vector Routing (DSDV) for Mobile Computers. ACM Special Interest Group on Data Communications (SIGCOMM). Ross, P.J. 1996. Taguchi Techniques for Quality Engineering. 2nd ed., New York: McGraw-Hill Inc. Roy, R.K. 2001. Design of Experiment Using the Taguchi Approach: 16 Steps to Product and Process Improvement. New York: Wiley. Sarahintu, M., Lee, M.H, & Mohamed, H. 2007. Determining the Effects of Scenario Metrics on the Performance of Dynamic Source Routing Using Taguchi Approach. Journal of Matematika 23(2): 121-132. Hazura Mohamed* Department of Industrial Computing Faculty of Information Science and Technology Universiti Kebangsaan Malaysia 43600 UKM Bangi, Selangor D.E.
MALAYSIA

Abolhasan, M., Wysocki, T. & Dutkiewicz, E. 2004. A Review of Routing Protocols for Mobile Ad Hoc. Elsevier Journal of Ad Hoc Network 2(1): 1-22. Ari,I., Jethani, N., Rangnekar, A., & Natarajan, S. 2000. Performance Analysis and Comparison of Ad Hoc Routing Protocols. Project Report. Boukerche, A. 2004. Performance Evaluation of Routing Protocols for Ad Hoc Wireless Networks. Journal Mobile Networks and Applications 9(4): 333-342. Das, S.R., Castaneda, R. & Yan, J. 2000. Simulation Based Performance Evaluation of Routing Protocols for Mobile Ad Hoc Networks. Journal Mobile Networks and Applications 5: 179-189. Kurkowski,S., Camp, T. & Colagrosso, M. 2005. MANET Simulation Studies: The incredibles. ACMs Mobile Computing and Communications Review 9(4): 50-61. Lakshmi, M. & Sankaranarayanan, P.E. 2005. Performance Analysis of Three Routing Protocols in Mobile Ad Hoc

Muhammad Hisyam Lee, Mazalan Sarahintu, Shaharuddin Salleh & Bahrom Sanugi Department of Mathematics, Faculty of Science Universiti Teknologi Malaysia 81310 Skudai, Johor Darul Takzim
MALAYSIA

*Corresponding author, email: hazura@ftsm.ukm.my Received: 19 May 2008 Accepted: 9 September 2008

Вам также может понравиться