Analysis and verification of network profile

Journal of Systems Engineering and Electronics Vol. 21, No.5, October 2010, pp.784-790Available online at www.jseepub.comAnalysis and verification of network profileWeiwei Chen!.*, Ning Huang', Yuqing Liu2, Ye Wang', and Rui Kang'1. School of Reliability and Systems Engineering, Beihang University, Bejing 100191, P. R. China;2. School of Computer Science and Engineering, Beihang University, Beijing 100191, P. R. China;3. Department of Information & Communication Engineering, Kongju National University, Chungnam 330717, Republic of KoreaAbstract: The elements of network profile are proposed. Basedinput which can be shown as the application traffic [2,7-on the network taffi distribution model, the network profile in-11]. This profile can be used to evaluate the reliability ofcludes the application request rate, the branch transfer probability,network structure in spite of the user access software. Thethe ratio of application requests, and the probability distribution offunctions of the object being tested include fragmentationthe requested objects. Based on the evaluation method of networkperformance reliabilit, four simulation cases are constructed inand reassembly, datagram encapsulation, connection con-OPNET software, and the rsuts show the four elements of prfile trol, fow control, error handing and diagnostics, adress-have impacts on the network reliability.ing, multiplexing, etc [12].Keywords: system engineering, network profile, performance re-According to current military needs, the network profilelabilit, network simulation.in the top layer can be directly used to guide the practice ofheadquarters. Therefore, this profile will be first discussedDOI: 10.3969/jissn.1004-4132 :2010.05.011in this paper.The detailed researches on top layer profile are as fol-1. Introductionlows. Reference [4] defined the network profile as a prob-ability, which can be calculated by the accessing times ofof nodes (such as end node, swithed node), communica- the service in a unit time, according to the operation pro-tion links and network services [1]. It implements the in- file of sofware defined as a simple set of operations andformation exchange among users through the existing sig-their probabilities of occurrence [5]. However, it is onlynaling and protocols. In order to evaluate the network relia-for network service reliability evaluation, because it didbility, the accurate network profile needs to be constructed.not describe the behavior of the group behavior accessingAccording to [2], the military network profile is con-the service. Reference [6] proposed the profile for Ad-hocstructed by the experience and intuition of military agent.network which ilustrates two contents, the time interval ofWhereas, that network profile cannot guarantee the warfare each application request and the application type. Unfor-illumination, because of the lack of accurate, deliberate,tunately, the content of this profile is incomplete, becauseand scientific methods. However, network profile can bethere is no further discussion on the different request types.constructed on different network layers, each of which hasTo this end, the elements of network profile based onindependent function, and each profile is obtained through the traffic distribution model are proposed in this paper.the top-down mission decomposition.Furthermore, the work of simulation and performance re-There are some valuable researches related in this area.liability evaluation was done to evaluate and analyze theFirstly, network profile is regarded as a top layer inputimpact on the network reliability.which can be shown as user's behavior [3- -6]. It can be2. Content of network profileused to evaluate the reliability of the whole network in-cluding the user access software and network structure. Network profile can be considered as a group behavior ofThe function of the object being tested is providing mul-using different applications. As discussed in [4,13], thetiple network service including resource sharing and datanetwork failure is mainly caused by the traffic distributiontransmission [3].on the network. The relation between network profile andSecondly, network profile is regarded as the lower layertraffc distribution is that the traffic is generated by the uti-Manuscript received May 6, 2010.lization of the network, while the traffic distribution is the*Corresponding author.result of the network中国煤化工network.YHCNMH G785Weiwei Chen et al: Analysis and verification of network profile2.1 Network taffic distribution modelthe service address. The CIS structure service generallyThe trffic distribution has three aspects in [14], namely:includes a source address and a destination address, but(i) Space distribution. It is the communicaion probabil-for multi-service and a distributed- service model, wherethere are more than one source address and destination ad-ity distribution in space, and also it is defined as the dis-dress. For these types of service modes, the server whichtribution of tafic source and destination node. (i) Timethe responses come from should be cleared. Therefore,distribution. The time interval of the taffic input which isthe ppication branch should be defined in network proflerepresented as the probability in time. (ii) Packet formatto show the different server addresses. Consequently, theand size distribution.users seleet the branch randomly, so the probability showsThe space distribution can be cassified as follows:the branch that has been chosen.(a) The given destinatin adress. The packets gener-(b) The time distribution of the traffic maps to the net-ated in source node are transmitted to destination address.work profile. The time distribution can be described as(b) Uniform distribution. The packets generated by thethe taffic model including the packet arrival process andsource node are delivred to all the other nodes, like broad-the amount of taffics in a certain time. Reference [18]casting in the network.proved that the size distribution of requested object is the(c) Hot distribution. The packets generated by the sourcemain control variable for the taffc. References [19,20]node are transmitted to a given node, where packets ex-used the FARIMA model to simulate the application traf-change frequently compared to others.fic of variable-it-rate (VBR) video \vhich is generated by(d) Local ditribution. The packets generated by the users' requests with high frequency. Reference [21] usedsource node are only sent to the near node in the wholethe ON/OFF model to simulate the data traffic which isnetwork.generated by local area network, both of them proved the(e) Matrix transfer addresses distribution. The sourcerequest rate is also a key factor. In addition, an applicnode and destination node are determined by transposedmay have various request types. Take FTP as an example,matrixesX and Y.this application has both“upload" and“download”types,The time distribution can be described as follows:different taffics in uplink and downlink will be generated(a) Prescribed time interval. The source node transmits by the different probabilistic combination of those applica-each packet after a prescribed period till the process is tion types. Therefore, the time distribution of the traffic canstopped.be rflected diferent application requests, the frequency of(b) Traffic model. The trfic model described in [15-requests, and the size of requested objects including down-17] includes ON/OFF model, FBMIFGN model, FARIMAload and upload objects in the user layer.model and M/G/∞model, respetively. The ON/OFF(c) The packet size maps to the network profile. Themodel ilustrates that the arriving sequence of packetssize of packet is decided not only by the network proto-s a random traffic decided by signaling generated ratio.cols, but also the size of requested files. For example, aThe FBMFGN model depicts that the tafic is a long-FTP download request may access to different files withrange dependent sequence of steady Gaussian noise. Indifferent size; a web browsing request may access to dif-the FARIMA model, the taffic is a steady reversible pro-ferent objects (pictures, plugins, and dynamic pages, etc.).cess sequence, and shows the long range and shor-rangeTherefore, the same request type may result in the ddependence under the different time scale. The M/G/oosize distribution of the responses, which makes the packetmodel describes a self: similar taf of which the packetsize also changes.arriving sequence is modeled as Poisson-distribution andFig.1 shows the traffic distribution model mapping tothe service time T is described as Geometric distribution.the network profile.The packet format is decided by the network type andthe packet size distribution which is efted by the size ofBranch transferprobabilitysending files and the network protocols.Space distributionApplication request2.2 Analysis of network profileThe distribution model of the network trfic maps to theTime distributionApplication requcstnetwork profile which is represented as the user's behavior,ratiothe following is a detailed discussion about the mapping.Packet sizeProbability distribution(a) The spatial distribution of the trafic maps to the net-of requested objectswork profile. The source address and destination address inFig. 1 Traffic distril中国煤化工“poilethe rff dstribution model map to the user's adres andTYHCNMHG.786Journal of Systems Engineering and Electronics Vol. 21, No.5, October 2010According to the above discussion, the network profileVoD on LAN1; VoD on LAN2; FTP on LAN3. The sizecan be defined as a set of the following four elements: Net-of VoD files which are deployed in servers A, B, and theworkProfile={(AppReqRatej, BranchTransProbj, AppRe-main server is 130 MB; the FTP service is also deployed inqRatioj, DisReqObjProbj),j= 1,2,3...,m}.the main server containing files A and B with size 1.5x 104The instruction is as follows:bytes and 2.9x 104 bytes, respectively.(a) AppReqRatej: The request rate of the application j.This parameter indicates the serial characteristic of the ap-_Linplication requests, and can be calculated by the numbers ofrequests per unit time.Main server(b) BranchTransProbj: The branch transfer probabilityimk 4of application j. This parameter denotes the applicationbranches in an application request serial. Users may ac-ickhone 明wcess different servers to obtain the network services whichDistributed server_ Amay have different application branches. Take the videoon demand (VoD) application as an example [22,23], itNink2 LinK3LinK9contains two servers: the distributed server and the mainLinly; Distributedserver_ Bserver. If the video on demand has been stored in thedistributed server, the video streaming will be transmittedby the scheduling strategy, which can be regarded as thebranch 1; Otherwise, the distributed server will send a fileFig. 2 Network modelmissing message to the main server and the video trans-mitted to the end users will be downloaded from the main4. Experiments and analysisserver, this is the application branch 2.(c) AppReqRatioj: Probabilistic combination of the dif-There are four experiments. In each experiment, the el-ferent request types of the application j. This parameterement value of network profile changes in order to studydenotes the ratio of different application types using thethe corresponding changes in network reliability. The sim-ulation time is set to 1 hour and the performance data areis the ratio of the numbers of“upload" to the“download".collcted every 5 s in each scenario. The data are processed(d) DisReqObjProbj: Probabilistic distribution of re-by the method which is described by Subsections 4.1 andquired objects of application j. This parameter indicates4.2.the probability of an object to be requested in the same ap-4.1 Evaluation method of the network performanceplication branch. For example, if there are 10“download"reliabilityFTP requests in a unit time, 2 of which access to the ob-ject A, the rest of which access to the object B. Then thelife data. With the design, manufacturing technology ad-DisReqObjProb will be {20%, 80%}.vances and the quality of materials improving, the product3. Case studiesreliability is increasing, so it is hard to get enough failuredata in a relatively short time, and for the complex system,In order to validate the elements of profile indeed affectits degradation mechanism is dificult to determine. There-the network reliability, a case is constructed by running thefore, the traditional reliability theory may not suitable forFTP and VoD applications.The network model constructed by using the OPNETproposed using the performance degradation parameters tosimulation software [24] is shown in Fig. 2. It con-evaluate the reliability. This is because the performancetains 3 groups with each group has 2 computers, theparameters contain life information which can reflect thecomputer is modeled by the Ethernet.system reliability in the usage. In regard to this, three mainthe 4 routers are modeled by the Atm4. ethernet2 gtwy steps are constructed to calculate the network reliability asmodel; 9 customer/server access links are modeled by theshown in Fig. 3:10BaseT. duplex link model; the 4 switches are modeled bythe Atm8. Lcrossconn4 model; 9 backbone access links are [30], this parameter is characterized as the network per-modeled by ATM SONET. _OC1 link models; the 3 serversformance in time ti relative to the benchmark to;are modeled by the Ethernet Server_adv model.(b) IPI (to) is initially set to 1. If the sample value isThere are three applications running on the network: greater than 1, th中国煤化工1. Oth-YHCNM HGWeiwei Chen et al: Analysis and verification of network profile787erwise, the number of un-failure times plus 1. The total congestion does not occur until the request rate reaches 30.sampling time is n in the network working time;After that, with the increase of congestion, the network re-(C) Calculate the network reliability R by Nelson modelliability is monotonically decreasing. The result coincides[31].with the expected conclusion, if the background applica-tions run stably, the network reliability shows tendency toDefine the basicMetricSetsmonotonically decrease with the increase of request rateonce the value exceeds a certain threshold.Define the benchmark| performance sample set {m}Table 1 Experiment 1 designVariable(AppReqRateDefine the weight vector {G}ScenarioofL AN1's VoD)Invariablenit. rUnit: requsts/100 s| Calculate the net work(a) LAN1 runs VoD appli-performance sample set {Ao}cation, with the Branch-1(TransProb is {1:1}; .Calculate IP(t)14(b) LAN2 runs VoD applica-2(tion, with the AppReqRate isN、「 F ailure times2610 requests per 300 s, Branch-IP(t)≥1?3(TransProbis {1:1};plus 134(C) LAN3 runs FTP download36application, with the AppRe-Non-failure times plus 14(qRate is 10 requests per 5 s,上Nand the DisReqObjProb isi+1≤n46{50%, 50%}.5(IN7(Calculatee R witlNelson10Fig. 3 Detained steps of network reliability evaluation1.4.2 Parameters of reliability evaluation model白0.8-The network basicMetricSet ={ downlink_ queuing de-g 0.60.4sponse Time, FTP Response Time }. The reasons of choos-0.ing the downlink queuing delay of Link 1 - Link 9 are026 10142026 30 34 36 40 44 46 50 70100as follows: () The nine links are the necessary routes forLAN1-LAN3 to access network service; (i) Traffic gener-Fig. 4 The impact of request rate on network rlibilityated by the VoD and FTP applications mainly run on thosedownlinks which makes their capacities are occupied. TheExperiment 2 The impact of branch transfer proba-response time of VoD and FTP is also an important factorbility on network reliability.to evaluate the network reliability from the user's point ofSimulation is configured with 11 scenarios which shownview.in Table 2. The branch 1 and branch 2 in this experimentLet the benchmark performance sample {mo} =refer to the application route from LAN1 to distribution{0.001 2 s, 8.56x10-6 s, 8.56x10-6 s, 8.56x10-6 s9x10-4 s, 8.56x10-6 s, 8.56x10-6 s, 0.001 2 s，sults in Fig. 5, the network reliability in scenariol and sce-8.56x10-6 s，14 s, 0.218 s}， weight vector {G} =nario11 is evidently lower than other scenarios in which{0.1, 0.05, 0.05, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1},the trend of reliability is non-monotonic. The curve showsthen{Ao}= {833, 1.17x105, 1.17x10, 1.17x105, 1.1xthat requests cause the traffic occupies one branch which103, 1.17x10, 1.17x 10, 833, 1.17x 105, 0.071, 4.59}.makes the load of links in this branch increase, result-ing in network congestion with the increase of retransmis-4.3 Experiment and analysission and link queuing delay, eventually leading to changesExperiment 1 The impact of the application requestin network reliability. Therefore, it is concluded that therate on the network reliability.branch transfer probability makes an impact on the net-Simulation is configured with 15 scenarios which shownwork reliability, which coincides with the expected con-in Table 1. The results are shown in Fig. 4. It shows that clusion. Meanwhile,中国煤化工paper hasMYHCNMH G788Journal of Systems Engineering and Electronics Vol. 21, No. 5, October 2010only two branches; we can deduce that if two branches lead of response file gradually become larger, thus more likelyto changes in network reliability, so do more branches.to cause network congestion, eventually network reliabilitydecreases. This is consistent with the desired conclusion:Table 2 Experiment 2 designTable 3 Experiment 3 designvTamsP,(BranchTransProb ofVariablScenarioLANI's VoD)Invariable(AppReqRatio of{branch1: branch2}LAN1's FTP)Unit: numbers of{download: upload}requests{10:0}(a) LANI runs VoD application, withAppReqRate is 10 requests per 100 s;{9:1}{ 10:0}(a) LAN3 runs both FTP upload(b) LAN2 runs VoD application, with{8:2}the AppReqRate is 10 requests per 100and download application, with Ap-{7:3}s, and BranchTransProb is {1:1};pReqRate is 10 requests per 5 s, nobranches, the the DisReqObjProb{6:4}(c) LAN3 runs FTP download ap-{7:3is {50%. 50%}:plication, with the AppReqRate is{5:5}(b) LAN1 runs VoD application,2 requests per 5 s, no branch, the{4:.6}DisReqObjProb is {50%, 50%}.with AppReqRate is 10 requests per300s. BranchTransProbis {1:1}:{3:7}4:6}(C) LAN2 run VoD application, as1on, d2the same settings withI ,AN1.{1:9}2:8}{0:10}1.05首0.95-10.90首0.50.850.8010/09/1 8/2 7/36/4 5/5 4/6 3/7 2/8 1/90/100.7 ↓BranchTransProbFig. 5 Impact of branch transfer probability on network reliability. 10/09/1 8/2 7/36/4 5/5 4/63/7 2/8 1/90/10AppReqRatioExperiment 3 The impact of the application requestFig. 6 Impact of application request ratio on network reliabilityratio on the network reliabilityAccording to the settings in Table 3, the simulation isTable4 Experiment 4 designconfigured to obtain the results as shown in Fig.6. TheVariableresult displayed is similar with Experiment 2. This curve(DisReqObjProbofresults from the reasons that the requests cause the traf-LAN3's FTP) {file A,file B} Unit:fic concentrated in uplink or downlink, and result in theprobability of objectnetwork congestion, as well as the decrease of network re-being requestedliability. In conclusion, the result is correspondent with the{100%，叮(a) LAN3 runs FTP downloadexpected: Changes in the application request Ratio will re-{90%, 10%}application，AppReqRate is 1sult in changes of the network reliability.requests per 5 s, no branch;Experiment 4 The impact of the probability distribu-{80%, 20%}(b) LANI runs VoD application,{70%, 30%}the AppReqRate is 10 requests pertion of the requested objects on the network reliability{60%, 40%}300 s, BranchTransProb is {1:1};11 simulation scenarios are configured according to Ta-{50%, 50%}(C) LAN2 runs VoD application, asble 4. The result in Fig. 7 shows that when the probabilitythe same setting as LAN1.{40%, 60%}distribution is from {1, 0} to {0.6, 0.4}, its reliability is 1,{30%, 70%}and its reliability decreases from {0.6, 0.4} to {0, 1}. The{20%, 80%}reason for this curve is: the size of fle B is larger than file1({10%, 90%}A, when the number of times that“download" requests ac-{0, 100cess to the file B are more than file A, the number and size中国煤化工-°THCNMH GWeiwei Chen et al: Analysis and verifcation of network profile7892051. (in Chinese)[5] J. D. Musa. Operational profiles in software reliability engi-白0.9neering. IEEE Software, 1993, 10(2): 13-32.吾0.8[6] C. Yang, Z. F. Zhu, W. Huang. Application of simulation tech-nology in reliability measure of Ad Hoc network. Proc. of豆0.7the 8th International Conference on Reliability, Maintainabil-0.6ity and Safety, 2009: 1137-1 140.[7] F. He. The performance management system of the CERNETbackbone based on the network trffic er，Beijing:ghua University, 2001. (in Chinese)DisReqObjProb[8] T. He, H Zhang, z. C. Li. A methodology for analyzing back-bone network trafic at stream-level. Proc. of CommunicationFig. 7 Impact of the probility distribution of the requested ob-jects on network reliability9] C. Fraleigh, s. Moon, B. Lyles, et al. Packet-level taffc mea-surements from the sprint IP backbone. IEEE Trans. on Net-he probability distribution of the requested objectsworks, 2003, 17(6): 6-16.changes result in the changes of network reliability.[10] C. Barakat, P. Thiran, G. Lannaccone, et al. Modeling Inter-net backbone trafic at the flow-level. IEEE Trans. on Signal .5. ConclusionsProcessing, 2003, 51(8): 2111-2124.The application request rate, the application request ratio,11] P. Liu, F. Liu, Z. M. Lei. Model of network tafic based onnetwork applications and network users. Proc. of Intermnationalthe branch transfer probability, and the probability distri-Symposium on Computebution of the requested objects are proposed as the profileSposin on Compuer Scene and Cmpuarional Tchnolelements in this paper. In order to prove that the four el-12] A. Leon-Garcia, 1. Widjaja. Communication networks: funda-ements have impact on the network reliability, four differ-mental concepts and key architectures. New York: McGraw-Hill Companies, 2003.ent experiments are designed based on the simulation and[13] L. L. Jiang. Research and implementation of the critical tech-evaluation method of network performance reliability. Thenique in large- scale network trffic measurement. Universityresults prove the expected conclusion. The following con-of Electronics Science and Technology of China, 2009. (inclusions are also drawn from this paper: () The networkChinese)reliability decreases with the increase of request rate af-[14] G. F. Han, H. M. Du, L. Jiang, et al. Construct NoC valida-ter the request rate exceed the threshold; (i) The networktion platform based on FPGA. Electronic Design Engineering,2010, 18(3): 90- 93.reliability will reach its minimum value when The request[15] W. Leland, M. Taqqu, W. Willinger, et al. On the self-similartype is all upload or download; (i) The network reliabilitynature of Ethernet traffic (extended version). IEEE/ACMwill reach its minimum value when all application requestsTrans. on Networking, 1994, 2(1): 1-15.take the same branch; (iv) The network reliability will de-16] B. Ryu, s. Lowen. Fractal taffic models for internet simula-tion. Proc. of ISCC Fifth IEEE Symposium on Computers andcreases when the request access to a larger size of file.Communications, Antibes-Juan les Pins, France, 2000: 200-The further research includes: (i) Nonlinear analysis be-tween network reliability and application request types, the[17] A. Nogueira, P. Salvador, R. Valadas. Modeling network traf-network reliability and the branch transfer probability. (i)fic with multi fractal behavior. 10th Intermational ConferenceImpact of the application sequential logic relationship onon Telecommunications, 2003: 1071-1077.the network reliability. (ii) Decomposition methods of18] K. Park, G. Kim, M. Crovella. On the efect of self- similarityon network performance. Proc. of the SPIE International Con-the network mission system based on the different networkference on Performance and Control of "Network System, 1997:layers. .296- -31019] J. Beran, R. Sherman, M. S. Taqqu, et al. Long-range depen-Referencesdence in variable bit rate video traffic. IEEE Trans. on Com-munications, 1995, 43(2/3/4): 1566- -1579.[1] M. Mitchell. Complex systems: network thinking. Artificial[20] P. Droz, J. Y. Le Boudec. A hig-speed sel-simiar ATM VBRIntelligence, 2006, 170: 1194-1212.taffic generator. IEEE GLOBECOM, 1996: 586- -590.[2] J. Y. Zhong, L. X. Yan, Y. Sheng. Study on the traffic mod-[21] J. w. Wei. Researches on fractal-based network trffic mod-eling and simulation method of the military communicationeling and queuing performance. Institute of Information En-network. System Engineering and Electronics, 2008, 30(9):gineering PLA Information Engineering University, 2006. (in1700-1703.(in Chinese)[3] R. Y. Li. Research on network reliability research of networkreliability evaluation methods. Beijing: Beihang University,22] S. H. G. Chan, F. Tobagi. Distributed servers architecture fornetworked video services. IEEE/ACM Trans. on Networking,[4] C. G. Bai, L. Su, Y. C. Zhao, et al. Is the reliability of web ser-2001, 9(2): 125-136.[23] F. Lin, C. Y. Zheng， X. WanVOD service model invices related to the change rate of operational profiles. Journalof Computer Research and Development, 2008, 45(12): 2044-broadband inform中国煤化工: in China. .TYHCNM HG790Journal of Systems Engineering and Electronics Vol. 21, No.5, October 2010Computer Applications and Sofware, 2008, 25(1): 20- -22.Ning Huang was borm in 1968. She is a professorin College of Reliability and Systems Engineer-[24] Application model user guide. OPNET 10.0 Product Docu-ing, Beihang University. She obtained her Ph.D.mentation, Inc, 2003.degree in 1997 from the same university. She has[25] J. Y. Zhao. Study on reliability modeling and applicationsbeen a visiting scholar in the University of Ili-based on performance degradation. National University of De-fense Technology, 2005. (in Chinese)nois Urbana-Champaign from 2007 to 2008. She261] E. L. Meyer, E. E. van Dyk. Assessing the reliability andis the author of more than forty scholarly papersdegradation of photovoltaic module performance parameters.covering various areas of network reliability andIEEE Trans. on Reliability, 2004, 53(1): 83- 92.software reliability. Her main research interests include network re-[27] J. Lee. Measurement of machine performance degradation us-liability and network profileing a neural network model. Computers in Industry, 1996,E-mail: hn@ buaa.edu.cn30(3);: 193 209.[28] w. Ke, C. Ren, K. Jin, et al. System performance, degradation,Yuqing Liu was borm in 1986. She is pursuingand reliability asessment. IEEE International Conference onher M.S. degree from College of Computer Sci-Industrial Engineering and Engineering Management, 2007:ence and Engineering, Beihang University. Her1216- 1220.research interests are network reliability and soft-[29] H. Lu, W. Kolarik, S. Lu. Real-time performance reliabilityware engineering.prediction. IEEE Trans. on Reliability, 2001, 50(4): 353- -357.E-mail:yuqingsunny@ gmail.com30] X. P. Jiang. Design and realization of an integrated evaluationmethod of network performance. Journal of Naval Universityof Engineering, 2006, 18(5): 75- -78. (in Chinese)31] J. Tian, S. Rudraraju, L. Zhao. Evaluating web software relia-bility based on workload and failure data extracted from serverYe Wang has been worked for three years inlogs. IEEE Trans. on Software Engineering, 2004, 30(11):Zhuhai Wanlida Electric Co., LtD.754- -769.ceived his M. s. in Information and Communi-Biographiesversity in Kongju, Korea in 2009. He held aPh. D. degree from the same university. His mainWeiwei Chen was born in 1982. She receivedresearch interests include mobile Ad Hoc net-her M.S. degree in College of Reliability andworks, wireless communication networks, andSystems Engineering, Beihang University inNGN mobiliy.006. She has been a visiting scholar in DeE-mail: wychina 2007 @ hotmail.compartment of Mechanical Engineering, Univer-sity of Maryland, USA, from 2007 to 2008. SheRui Kang was born in 1966 and received his B.is pursuing her Ph.D. degree from College ofS. from Beihang University in 1990. He is theReliability and Systems Engineering, BeihangUniversity. Her main research interests include network reliabilitychair professor in College of Reliability and Sys-tems Engineering, Beihang University. He wasand network profile.a visiting professor of more than four universi-E-mail: Rainychen @ dse.buaa.edu.cnties. His main research interests include reliabil-ity design and experiment technology based onthe phhysics of failure (PoF), tnostic andhealth management (PHM), logistics support, and network reliability.E-mail: kangrui @ buaa.edu.cn中国煤化工MHCNM HG