- Email: [email protected]
Context Awareness Based Bandwidth Management Scheme for Ad hoc Network
Available online at www.sciencedirect.com
ScienceDirect Procedia Computer Science 107 (2017) 484 – 489
International Congress of Information and Communication Technology (ICICT 2017)
Context Awareness based Bandwidth Management Scheme for Ad hoc Network Wang Haitaoa,*, Song Lihuab, Chen Huia, Yan Lia, Zhang Guominb, Hu Qianga a Information Management Center, PLA Univ. of Sci. & Tech.ˈNanjing, China College of Command Information System, PLA Univ. of Sci. & Tech.ˈNanjing, China * Corresponding author: [email protected]
b
Abstract In order to guarantee key businesses when bandwidth resource is in shortage, an adaptive and flexible Context-Aware Bandwidth Management Scheme (CABMS) is proposed to improve network survivability. Nodes firstly query the local context information, and use Bayesian Network (BN) to determine the importance of current business, further to determine the utility function of bandwidth allocation. Through establishing the dual problem of original one, the "shadow price" of bandwidth is introduced, so that the nodes are able to adjust bandwidth requests on their own according to the price, with the convergence of allocation result. In the CABMS business is classified into different levels. When bandwidth resources is in shortage, the high-class business will be biased in bandwidth allocation; when in severe shortage, some regular business bandwidth requests will be rejected, in order to guarantee the bandwidth requests of key business. Simulation results indicate that CABMS can assign more bandwidth to relatively important business under given conditions, compared with proportional fairness. Specifically, urgent business bandwidth allocation increases by about 42%, important business bandwidth allocation remains essentially unchanged and the regular allocate bandwidth falls by about 37%. Keywords: Ad hoc Network; Context Awareness; Bandwidth Management; Bayes Network; Proportional Fairness
1. Introduction Because of the flexibility and rapidness of constructing network on various occasions, Ad hoc networks are widely used in emergency situations and in the battlefield environments. Bandwidth resource in Ad hoc network is scarce, when many data flows in the network contend for bandwidth; reasonable bandwidth allocation methods are needed. Currently, many research works have been done on bandwidth management for Ad hoc networks. Fang Z Y proposes two bandwidth allocation schemes based on gaming theory and they can change the utility function to balance the fairness and efficiency of bandwidth allocation [1]. Xue Y introduced the concept of shadow price of the maximum clique and a price based bandwidth allocation algorithm is proposed to achieve the utilization sum maximization of data flows with certain fairness [2]. In references [3][4], an Ad hoc bandwidth allocation algorithm
1877-0509 © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of the 7th International Congress of Information and Communication Technology doi:10.1016/j.procs.2017.03.094
485
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489
based on auction mechanism is proposed, data flows determine current bandwidth price according to the budget and compete for resources, reducing the complexity and accelerating the convergence time. In reference [5], a distributed admission control algorithm for Ad hoc networks is put forward which does not need knowing the exact amount of remaining bandwidth. However, these studies don’t consider the rational allocation of bandwidth aiming to specific battle scenarios. In fact, the importance of different service follows in practical network environment varies greatly. From the perspective of network survivability, when bandwidth is scarce, more bandwidth should be allocated to the more important data flow, thus ensuring the completion of key services. In addition, the existing Ad hoc network bandwidth management solutions pay little attention to network context of applications, including network heterogeneity and special user needs. Anind Dey gave a more generic definition of context: it is any information that can be used to characterize an entity in its current state, the entity can be a person, thing or any other objects interacting with the user and application, including the user and the application itself [6]. In real network environment relevant context data can be collected by various sensors and detectors [7], and then tools such as Bayesian Networks are used to model and reason about context. Finally, context data will be stored in the repository for later query and usage. Bayesian Networks belongs to the probabilistic graphical model which is suitable for model and analysis of uncertainty problems and it has unique advantages in dealing with uncertainty and provides efficient inference algorithms [8]. At present, as an context inference toll Bayesian Network has been widely recognized by people [9-10]. Based on context-aware applications can timely learn environmental information to make reasonable actions and provide relevant information or services to users, allowing users getting satisfactory services efficiently at a lower cost. 2. Descriptions of CABMS 2.1. Context Awareness and Decision-making To achieve adaptive allocation of bandwidth based on service importance, context-aware based Ad hoc network bandwidth management mechanisms (CABMS) requires context reasoning technologies to reason about the importance of various services. Specifically, CABMS uses Bayesian network as inference tool, which can express uncertainty in a visual way beneficial to understanding context model. Factors influencing service importance are various and complex. To simplify analysis, taking battlefield environment for an example, using seven kinds of common context information to evaluate the importance of services, including service type (Business), user identity (Identity), fighting state (Fight), environment noise (Noise), user acceleration (Acceleration), mechanical vibration frequency (Vibration) and service significance (Significance). Additionally, assuming these factors are discrete variables. Depending on whether the variable can be observed, those seven factors can be divided into observable variables Vobserved={I, N, A, V, B} and hidden variables Vhidden={F,S}. According to the causality of the impact of context on service importance, Bayesian Network is constructed. In practical applications parameters of Bayesian Network can be trained based on a large number of samples. When no data is missing the maximum likelihood method is used to estimate the parameters. When data is missing, EM algorithm is utilized for parameter learning. After gaining context, using structured Bayesian Network to inference about relevant information. Specially, CABMS uses the clique tree method to calculate the probability distribution. Clique tree is an exact infere1nce algorithm with high computing speed and its main step is to convert the Bayesian Network to clique tree and calculate the probability by belief propagation [13]. 2.2. The Procedure of Bandwidth Allocation To reflect impacts of different service importance on bandwidth allocation, Sigmoid function is used by CABMS to indicate the utility of data flows, depicted as follows. °U i ( xi ) ° ®U i ( xi ) ° °U i ( xi ) ¯
i 0, if xi Rmin i 1, if xi t Rmax
2 1, M ! 0, otherwise 1 e M x i
(1)
486
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489
Thereinto, M is utility gain factor and its value is [0,1]. M can control the utility of data flows to allocate more i i bandwidth to the data flow with bigger M . Obviously, when xi ª¬Rmin , Rmax º¼ U i ( xi ) is a strict concave function. m
¦J
m
j
j 1
Aj ,i represents the path cost of data flow j. Let PP= ¦ J j Aj ,i and z
eM x , the following formula can be i
j 1
obtained: 2M · § z2 ¨ 2 ¸ z 1 0 PP ¹ ©
(2)
The results are: ° z1 ° ® ° ° z2 ¯
M
1
PP
M PP
1
M § M · 2¸ ¨ PP © PP ¹
(3)
M § M
· 2¸ ¨ PP © PP ¹
M § M · M 2 ¸ t 0 , so ! 2 and 0 z1 d 1 , z2 t 1 DŽSince 0 z d 1 , only if z1 is ¨ PP © PP PP ¹ solvable, after introducing z eM x , we can obtain: Since M ! 0 , PP ! 0 and
i
xi*
§ M M § M ·· 1 2 ¸ ¸ M ln ¨ ¨ ¨ PP PP PP © ¹ ¸¹ ©
(4)
Here the service imporatance index is N ! 0 , and M (2 N ) PP So we can obtain: ln N N N 2 1 xi*
2 N PP
(5)
(6)
The rule indicates that when path cost is fixed, service importance is increased by one level and its allocated bandwidth increases 50%. Let xEM , xIM and xGE express the allocated bandwidth of emergency service, important service and general service respectively, the following formula can be obtained: xEM xIM 3 2 , when PPEM PPIM PPGE (7) ® ¯ xIM xGE 3 2
For three kinds of service, N EM =1, N IM =0.22 and N GE =0.083. To simplify the process, the node with the smallest distance sum to all other nodes in the clique is selected as the clique head node. The clique head node is responsible for collecting the bandwidth requests of service flows passing the clique and updating bandwidth price of the clique. Adopting gradient projection algorithm (GPA) in reference [1], after receiving bandwidth requests of service flows, the new price of bandwidth is calculated according to the following formula. ª
§ ·º (8) © ¹¼ ¬ In formula (8), [a]+ indicates max{0ˈa} and E is setp size. When E is sufficiently small the resultsof bandwith allocation will convergence. It can be seen that when bandwidth requirement is more than clique capacity, the bandwidth price will increase; otherwise, it will decrease. The clique head node will inform new price to all * data flows passing that clique, and then data flows calculate the optimal bandwidth allocation value x (r 1) . Next, req nodes determine actual bandwidth requirements accordign to foumula (9) and return x to the clique head node and then start next round of price calculation. It performs such iterations untill bandwidth request values of data flows convergence and set as fianl the results of bandwidth allocation.
J i (r 1) «J i (r ) E ¨ ¦ x j Ai , j Ci ¸ » j
xreq
min max x* , Rmin , Rmax
(9)
In practice, when allocated bandwidth exceeds the maximum bandwidth requirementsof data flows, more bandwidth will no longer increase its utility, but wasting scare bandwidth. When bandwidth quota falls below the
487
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489
minimum bandwidth requirement its utility value is degraded to zero and also wastes bandwidth. CABMS considers the maximum and minimum bandwidth requirements of data flows, thus improving the efficiency of bandwidth usage. Specially, when bandwidth is sufficient, the maximum bandwidth demand values are allocated for all data flows; when the clique capacity is insufficient to meet the minimum bandwidth requirement of all data flows, certain non-critical services should be rejected based on service importanceto guarantee the successful completion of critical services. When receiving requested bandwidth of data flows, the clique head node returns the calculated bandwidth prices and residual capacity to these data flows. Then, data flows determine whether the current residual bandwidth can meet their minimum needs. If so, it will continue to participate in the bandwidth allocation process; otherwise, it will exit the bandwidth allocation process. If the bandwidth capacity is insufficient to meet all bandwidth requests, the clique head node sends a bandwidth shortage message Requst_to_abort to all contending data flows to remind partial data flows cancelling or postponing the bandwidth request process. After receiving the Requst_to_abort message, data flows will calculate the corresponding probability to exit bandwidth competition according to the importance of its service. Selective exit policy can imporve the fairness to avoid starving service flows with lower importance. Formula (10) gives the expression selective exit probability. °Pr Abort w1 f min_rate w2 f importance ° 2 K R R ® f min_rate 1 1 e ° ° f importance 1 1 e K I I 2 ¯
MAX min
min
MAX
(10)
In formula (10), f min_rate and fimportance indicates the minimum bandwidth requirements proportion function and the importance proportion function repectively. Here, f min_rate and fimportance >0,1@ . w1 and w2 are allocation weights and w1 w2 1 . In particularly, w1 0.1 and w2 0.9 . K is a constant and when K 2 obtained by simulation, the MAX ideal bandwidth allocation results can be obtained. Rmin is the minimum bandwidth requirement limit. I ^1, 2,3` ,
I MAX 3 , and the smaller I is and the higher the service importance. When service importance and the minimum bandwidth requirements are larger, exiting probabilities are higher. In addition, upon receiving the message Request_to_abort multiple data flows mya exit bandwidh competition process at the same time. In this case after the residula data flows obtained the required bandwidth, there may still be some bandwidth for exited data flows. For utilizing bandwidht sufficiently, the temporary exitted data flows will compete the residual bandwidth in the same above procedure once again, until all data flows obtain corresponding bandwidth or exit the competition entirely. 3. Simulation and Analysis 3.1. Simulation Environment Setting In simulation experiment a Ad hoc network including 10 nodes is deployed in the area of 1000h1000 m2, as shown in figure 2. The coverage of each node is a circle with radius of 400 m. In figure 2, the source nodes are {5,1,3} and the corresponding destination nodes are {8,10,9}. The routing algorithm is Floyd algorithm can the routing paths of three data flows can be obtained: F1={1,4,6,10}, F3={3,6,10,9} and F5={5,4,6,8}. According to active links clique resources can be bulit. Given the network topology and the paths, the largest cliques of Q1 and Q2 are depicted in figure 3, in which Q1={(1,4), (5,4), (4,6), (3,6), (6,10), (6,8),} and Q2={(9,10), (4,6) and (3,6), (6,10), (6,8)}. k indicates the importance level of service and N is the maximum number of importance level. Here, N=3 and bigger k means more important service. Fairness index is defined form the perspective of service importance and the bandwidth assigned to services should be propotional to the importance of services.
488
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489 km 1
5
0.9 0.8
9 7
1
0.6
10 2
0.3
0
4
10 6
6
0.2
8
Data flow 1
8
3
Data flow 3
0.1
3 0
0.1
0.2
0.3
0.4
Q2
1
4
0.5 0.4
9
5
Q1
0.7
Data flow 5
0.5
0.6
0.7
0.8
0.9
1
km
Fig. 2. network topology generated randomly
Fig. 3. the maximum cliques Q1 and Q2
3.2. Analysis of Simulation Results Assuming that the minimum bandwidth requirement of data flows is zero and the maximum bandwidth requirement is 3Mbps. Bandwidth allocations under CABMS and proportional fairness criterion are given in figure 4 and figure 5 respectively. Bandwidth allocation results for three data flows under CABMS are: {x1=1.5790, x2=1.0481, x3=1.0559}. Under the proportional fairness criterion bandwidth allocation results are: {x1=1.1112, x2=1.6668, x3=1.1112}. Mbps 33 2. 5
2. 5
22 1. 5
1.5
22 1. 5
1.5
11
11 0. 50 0
0.5
Node 㢲⛩1 Node 1 㢲⛩5 Node 5 㢲⛩3 3
2.5
࠶Bandwidth 䝽 ᑖ ᇭ (Mbps)
2.5
Bandwidth ࠶ 䝽 ᑖ ᇭ (Mbps)
Mbps 33
Node 㢲⛩1 Node 1 㢲⛩5 Node 5 㢲⛩3 3
2 20 0
4 40 0
6 60
8 80 0Rounds 䖞⅑ 0
10 100
0
12 120
0. 50 0
14 140
0
Fig. 4. Bandwidth allocations under CABMS
P ri c e
0. 0.1 1 0.0 0.09 9 0.0 0.08 8 0.0 0.07 7 0.0 0.06 6 0.0 0.05 5 0.0 0.04 4 0.0 0.03 3 0.0 0.02 2 0.0 0.01 1 0 0 0 0
1 0
0.5
0
2 0
10
3
20
30 Round 0 䖞 ⅑
s
4 0
40
5 0
50
6 0
60
Fig. 5. Bandwidth allocations under proportional fairness
?? 1 1 Clique Clique ? 2 2
2 20 0
4 40 0
6 60
0 Rounds
8 80 0
10 100
0
12 120 0
14 140
0
Fig. 6. Bandwidth prices under CABMS
In figure 4 and figure 5, bandwidth allocation values will convergence after some number of rounds. The bigger of bandwidth allocation step size is and the faster of convergence speed. Data flows F1 and F5 in Q1 and Q2 involve the same number of links. In accordance with rules of CABMS, bandwidth allocation of F1 is 1.5 times the bandwidth of F5, and the bandwidth of F1 is no more than 2.25 times the bandwidth of F3. This is because F3 in Q1
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489
489
only involves two links, in more competitive Q1, F3 pays lower price than the other two data flows to get the same bandwidth. So it can be allocated more bandwidth. Under the proportional fairness criterion due to occupation of less link resoruces of Q1, general service (F3) get more 50% bandwidth allocation than that of the other two data flows, which makes QoS of two data flows with higher service importance degrades. In the bandwidth allocation CABMS gives data flow (F1) with high importanc greater N value, thus obtaining more bandwidth than that of proportional fairness bandwidth allocation criterion. Obviously, CABMS gets better bandwidth allocation fairness. The changes of bandwidth prices under CABMS is shown in figure 6. As can be seen from the figure 6, since the capacity of Q1 is insufficient to fully meet the bandwidth requirements of the data flows, so bandwidth prices are increasing and eventually stabilizing. For Q2 due to initial larger bandwidth requirements of data flows the bandwidth prices start increasing. But with the changes of bandwidth requirements, the bandwidth prices will gradually be more reasonable. Then, bandwidth capacity of Q2 will fully meet bandwidth needs, and the bandwidth price will drop to zero. 4. Conclusions In practical network application environments, particularly under battlefield circumstances, the importance of various services is different. In bandwidth limited Ad hoc network environment, in order to guarantees successful completion of key services, a context-aware bandwidth allocation scheme named CABMS is proposed. CABMS takes advantage of various sensors to collect data as the basis for context reasoning. Bayesian Network is used as a reasoning tool to infer the importance of services. According to the importance degree of data flows competing scarce bandwidth resources more bandwidth will be allocated to the vital service and thus improving the network survivability. However, to simplify the analysis, in CABMS all variables are assumed to be discrete variables, but in reality some variables (such as noise) are continuous. In addition, in CABMS bandwidth allocation, the clique head node needs collecting bandwidth requirements of all data flows to calculate new bandwidth prices, requiring synchronization among data flows. All these problems are more difficult to solve in dynamic Ad hoc networks. Therefore, an asynchronous bandwidth allocation scheme can be considered for future. References 1. FANG Z, BENSAOU B. Fair bandwidth sharing algorithms based on game theory frameworks for wireless ad-hoc networks[C]//INFOCOM 2004: Twenty-third Annual Joint Conference of the IEEE Computer and Communications Societies. Hong Kong: IEEE, 2004: 1284-1295. 2. XUE Y, LI B, Nahrstedt K. Price-based Resource Allocation in wireless ad hoc networks[C]. Eleventh International Workshop on Quality of Service, Monterey, CA, June, 2003:79-96. 3. CURESCU C, NADJM-TEHRANI S. A Bidding Algorithm for Optimized Utility-Based Resource Allocation in Ad hoc Networks[J]. Mobile Computing IEEE Transactions on, 2008, 7(12):1397-1414. 4. KAO B, LEE L, K. ROBERT LAI. Multi-hop Auction-Based Bandwidth Allocation in Wireless Ad hoc Networks[C] //IEEE International Conference on Advanced Information Networking & Applications. Singapore: IEEE, 2011:772-778. 5. Zhao H, Garcia-Palacios E, Wei J, et al. Distributed resource management and admission control in wireless ad hoc networks: a practical approach[J]. Iet Communications, 2012, 6(8):883-888. 6. DEY AK. Understanding and using context[J]. Personal and ubiquitous computing, 2001, 5(1): 4-7. 7. PERERA C, ZASLAVSKY A, CHRISTEN P, et al. Context aware computing for the internet of things: A survey[J]. Communications Surveys & Tutorials, IEEE, 2014, 16(1): 414-454. 8. WITZIG T, ZOLLNER J M, PANGERCIC D, et al. Context aware shared autonomy for robotic manipulation tasks[C]//Intelligent Robots and Systems (IROS), 2013 IEEE/RSJ International Conference on. Tokyo: IEEE, 2013: 5686-5693. 9. KO K E, SIM K B. Development of context aware system based on Bayesian network driven context reasoning method and ontology context modeling[C]//Control, Automation and Systems, 2008. ICCAS 2008. International Conference on. Seoul: IEEE, 2008: 2309-2313. 10. RHO W H, CHO S B. Context-aware smartphone application category recommender system with modularized Bayesian networks[C]// IEEE Natural Computation (ICNC), Foz do Iguaçu, 2014: 775-779. 11. Anthony D. Wood, John A. Stankovic, Gilles Virone. Context-Aware Wireless Sensor Networks for Assisted Living and Residential Monitoring[J], IEEE Network, 2008,22(4):26-33.
ScienceDirect Procedia Computer Science 107 (2017) 484 – 489
International Congress of Information and Communication Technology (ICICT 2017)
Context Awareness based Bandwidth Management Scheme for Ad hoc Network Wang Haitaoa,*, Song Lihuab, Chen Huia, Yan Lia, Zhang Guominb, Hu Qianga a Information Management Center, PLA Univ. of Sci. & Tech.ˈNanjing, China College of Command Information System, PLA Univ. of Sci. & Tech.ˈNanjing, China * Corresponding author: [email protected]
b
Abstract In order to guarantee key businesses when bandwidth resource is in shortage, an adaptive and flexible Context-Aware Bandwidth Management Scheme (CABMS) is proposed to improve network survivability. Nodes firstly query the local context information, and use Bayesian Network (BN) to determine the importance of current business, further to determine the utility function of bandwidth allocation. Through establishing the dual problem of original one, the "shadow price" of bandwidth is introduced, so that the nodes are able to adjust bandwidth requests on their own according to the price, with the convergence of allocation result. In the CABMS business is classified into different levels. When bandwidth resources is in shortage, the high-class business will be biased in bandwidth allocation; when in severe shortage, some regular business bandwidth requests will be rejected, in order to guarantee the bandwidth requests of key business. Simulation results indicate that CABMS can assign more bandwidth to relatively important business under given conditions, compared with proportional fairness. Specifically, urgent business bandwidth allocation increases by about 42%, important business bandwidth allocation remains essentially unchanged and the regular allocate bandwidth falls by about 37%. Keywords: Ad hoc Network; Context Awareness; Bandwidth Management; Bayes Network; Proportional Fairness
1. Introduction Because of the flexibility and rapidness of constructing network on various occasions, Ad hoc networks are widely used in emergency situations and in the battlefield environments. Bandwidth resource in Ad hoc network is scarce, when many data flows in the network contend for bandwidth; reasonable bandwidth allocation methods are needed. Currently, many research works have been done on bandwidth management for Ad hoc networks. Fang Z Y proposes two bandwidth allocation schemes based on gaming theory and they can change the utility function to balance the fairness and efficiency of bandwidth allocation [1]. Xue Y introduced the concept of shadow price of the maximum clique and a price based bandwidth allocation algorithm is proposed to achieve the utilization sum maximization of data flows with certain fairness [2]. In references [3][4], an Ad hoc bandwidth allocation algorithm
1877-0509 © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of the 7th International Congress of Information and Communication Technology doi:10.1016/j.procs.2017.03.094
485
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489
based on auction mechanism is proposed, data flows determine current bandwidth price according to the budget and compete for resources, reducing the complexity and accelerating the convergence time. In reference [5], a distributed admission control algorithm for Ad hoc networks is put forward which does not need knowing the exact amount of remaining bandwidth. However, these studies don’t consider the rational allocation of bandwidth aiming to specific battle scenarios. In fact, the importance of different service follows in practical network environment varies greatly. From the perspective of network survivability, when bandwidth is scarce, more bandwidth should be allocated to the more important data flow, thus ensuring the completion of key services. In addition, the existing Ad hoc network bandwidth management solutions pay little attention to network context of applications, including network heterogeneity and special user needs. Anind Dey gave a more generic definition of context: it is any information that can be used to characterize an entity in its current state, the entity can be a person, thing or any other objects interacting with the user and application, including the user and the application itself [6]. In real network environment relevant context data can be collected by various sensors and detectors [7], and then tools such as Bayesian Networks are used to model and reason about context. Finally, context data will be stored in the repository for later query and usage. Bayesian Networks belongs to the probabilistic graphical model which is suitable for model and analysis of uncertainty problems and it has unique advantages in dealing with uncertainty and provides efficient inference algorithms [8]. At present, as an context inference toll Bayesian Network has been widely recognized by people [9-10]. Based on context-aware applications can timely learn environmental information to make reasonable actions and provide relevant information or services to users, allowing users getting satisfactory services efficiently at a lower cost. 2. Descriptions of CABMS 2.1. Context Awareness and Decision-making To achieve adaptive allocation of bandwidth based on service importance, context-aware based Ad hoc network bandwidth management mechanisms (CABMS) requires context reasoning technologies to reason about the importance of various services. Specifically, CABMS uses Bayesian network as inference tool, which can express uncertainty in a visual way beneficial to understanding context model. Factors influencing service importance are various and complex. To simplify analysis, taking battlefield environment for an example, using seven kinds of common context information to evaluate the importance of services, including service type (Business), user identity (Identity), fighting state (Fight), environment noise (Noise), user acceleration (Acceleration), mechanical vibration frequency (Vibration) and service significance (Significance). Additionally, assuming these factors are discrete variables. Depending on whether the variable can be observed, those seven factors can be divided into observable variables Vobserved={I, N, A, V, B} and hidden variables Vhidden={F,S}. According to the causality of the impact of context on service importance, Bayesian Network is constructed. In practical applications parameters of Bayesian Network can be trained based on a large number of samples. When no data is missing the maximum likelihood method is used to estimate the parameters. When data is missing, EM algorithm is utilized for parameter learning. After gaining context, using structured Bayesian Network to inference about relevant information. Specially, CABMS uses the clique tree method to calculate the probability distribution. Clique tree is an exact infere1nce algorithm with high computing speed and its main step is to convert the Bayesian Network to clique tree and calculate the probability by belief propagation [13]. 2.2. The Procedure of Bandwidth Allocation To reflect impacts of different service importance on bandwidth allocation, Sigmoid function is used by CABMS to indicate the utility of data flows, depicted as follows. °U i ( xi ) ° ®U i ( xi ) ° °U i ( xi ) ¯
i 0, if xi Rmin i 1, if xi t Rmax
2 1, M ! 0, otherwise 1 e M x i
(1)
486
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489
Thereinto, M is utility gain factor and its value is [0,1]. M can control the utility of data flows to allocate more i i bandwidth to the data flow with bigger M . Obviously, when xi ª¬Rmin , Rmax º¼ U i ( xi ) is a strict concave function. m
¦J
m
j
j 1
Aj ,i represents the path cost of data flow j. Let PP= ¦ J j Aj ,i and z
eM x , the following formula can be i
j 1
obtained: 2M · § z2 ¨ 2 ¸ z 1 0 PP ¹ ©
(2)
The results are: ° z1 ° ® ° ° z2 ¯
M
1
PP
M PP
1
M § M · 2¸ ¨ PP © PP ¹
(3)
M § M
· 2¸ ¨ PP © PP ¹
M § M · M 2 ¸ t 0 , so ! 2 and 0 z1 d 1 , z2 t 1 DŽSince 0 z d 1 , only if z1 is ¨ PP © PP PP ¹ solvable, after introducing z eM x , we can obtain: Since M ! 0 , PP ! 0 and
i
xi*
§ M M § M ·· 1 2 ¸ ¸ M ln ¨ ¨ ¨ PP PP PP © ¹ ¸¹ ©
(4)
Here the service imporatance index is N ! 0 , and M (2 N ) PP So we can obtain: ln N N N 2 1 xi*
2 N PP
(5)
(6)
The rule indicates that when path cost is fixed, service importance is increased by one level and its allocated bandwidth increases 50%. Let xEM , xIM and xGE express the allocated bandwidth of emergency service, important service and general service respectively, the following formula can be obtained: xEM xIM 3 2 , when PPEM PPIM PPGE (7) ® ¯ xIM xGE 3 2
For three kinds of service, N EM =1, N IM =0.22 and N GE =0.083. To simplify the process, the node with the smallest distance sum to all other nodes in the clique is selected as the clique head node. The clique head node is responsible for collecting the bandwidth requests of service flows passing the clique and updating bandwidth price of the clique. Adopting gradient projection algorithm (GPA) in reference [1], after receiving bandwidth requests of service flows, the new price of bandwidth is calculated according to the following formula. ª
§ ·º (8) © ¹¼ ¬ In formula (8), [a]+ indicates max{0ˈa} and E is setp size. When E is sufficiently small the resultsof bandwith allocation will convergence. It can be seen that when bandwidth requirement is more than clique capacity, the bandwidth price will increase; otherwise, it will decrease. The clique head node will inform new price to all * data flows passing that clique, and then data flows calculate the optimal bandwidth allocation value x (r 1) . Next, req nodes determine actual bandwidth requirements accordign to foumula (9) and return x to the clique head node and then start next round of price calculation. It performs such iterations untill bandwidth request values of data flows convergence and set as fianl the results of bandwidth allocation.
J i (r 1) «J i (r ) E ¨ ¦ x j Ai , j Ci ¸ » j
xreq
min max x* , Rmin , Rmax
(9)
In practice, when allocated bandwidth exceeds the maximum bandwidth requirementsof data flows, more bandwidth will no longer increase its utility, but wasting scare bandwidth. When bandwidth quota falls below the
487
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489
minimum bandwidth requirement its utility value is degraded to zero and also wastes bandwidth. CABMS considers the maximum and minimum bandwidth requirements of data flows, thus improving the efficiency of bandwidth usage. Specially, when bandwidth is sufficient, the maximum bandwidth demand values are allocated for all data flows; when the clique capacity is insufficient to meet the minimum bandwidth requirement of all data flows, certain non-critical services should be rejected based on service importanceto guarantee the successful completion of critical services. When receiving requested bandwidth of data flows, the clique head node returns the calculated bandwidth prices and residual capacity to these data flows. Then, data flows determine whether the current residual bandwidth can meet their minimum needs. If so, it will continue to participate in the bandwidth allocation process; otherwise, it will exit the bandwidth allocation process. If the bandwidth capacity is insufficient to meet all bandwidth requests, the clique head node sends a bandwidth shortage message Requst_to_abort to all contending data flows to remind partial data flows cancelling or postponing the bandwidth request process. After receiving the Requst_to_abort message, data flows will calculate the corresponding probability to exit bandwidth competition according to the importance of its service. Selective exit policy can imporve the fairness to avoid starving service flows with lower importance. Formula (10) gives the expression selective exit probability. °Pr Abort w1 f min_rate w2 f importance ° 2 K R R ® f min_rate 1 1 e ° ° f importance 1 1 e K I I 2 ¯
MAX min
min
MAX
(10)
In formula (10), f min_rate and fimportance indicates the minimum bandwidth requirements proportion function and the importance proportion function repectively. Here, f min_rate and fimportance >0,1@ . w1 and w2 are allocation weights and w1 w2 1 . In particularly, w1 0.1 and w2 0.9 . K is a constant and when K 2 obtained by simulation, the MAX ideal bandwidth allocation results can be obtained. Rmin is the minimum bandwidth requirement limit. I ^1, 2,3` ,
I MAX 3 , and the smaller I is and the higher the service importance. When service importance and the minimum bandwidth requirements are larger, exiting probabilities are higher. In addition, upon receiving the message Request_to_abort multiple data flows mya exit bandwidh competition process at the same time. In this case after the residula data flows obtained the required bandwidth, there may still be some bandwidth for exited data flows. For utilizing bandwidht sufficiently, the temporary exitted data flows will compete the residual bandwidth in the same above procedure once again, until all data flows obtain corresponding bandwidth or exit the competition entirely. 3. Simulation and Analysis 3.1. Simulation Environment Setting In simulation experiment a Ad hoc network including 10 nodes is deployed in the area of 1000h1000 m2, as shown in figure 2. The coverage of each node is a circle with radius of 400 m. In figure 2, the source nodes are {5,1,3} and the corresponding destination nodes are {8,10,9}. The routing algorithm is Floyd algorithm can the routing paths of three data flows can be obtained: F1={1,4,6,10}, F3={3,6,10,9} and F5={5,4,6,8}. According to active links clique resources can be bulit. Given the network topology and the paths, the largest cliques of Q1 and Q2 are depicted in figure 3, in which Q1={(1,4), (5,4), (4,6), (3,6), (6,10), (6,8),} and Q2={(9,10), (4,6) and (3,6), (6,10), (6,8)}. k indicates the importance level of service and N is the maximum number of importance level. Here, N=3 and bigger k means more important service. Fairness index is defined form the perspective of service importance and the bandwidth assigned to services should be propotional to the importance of services.
488
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489 km 1
5
0.9 0.8
9 7
1
0.6
10 2
0.3
0
4
10 6
6
0.2
8
Data flow 1
8
3
Data flow 3
0.1
3 0
0.1
0.2
0.3
0.4
Q2
1
4
0.5 0.4
9
5
Q1
0.7
Data flow 5
0.5
0.6
0.7
0.8
0.9
1
km
Fig. 2. network topology generated randomly
Fig. 3. the maximum cliques Q1 and Q2
3.2. Analysis of Simulation Results Assuming that the minimum bandwidth requirement of data flows is zero and the maximum bandwidth requirement is 3Mbps. Bandwidth allocations under CABMS and proportional fairness criterion are given in figure 4 and figure 5 respectively. Bandwidth allocation results for three data flows under CABMS are: {x1=1.5790, x2=1.0481, x3=1.0559}. Under the proportional fairness criterion bandwidth allocation results are: {x1=1.1112, x2=1.6668, x3=1.1112}. Mbps 33 2. 5
2. 5
22 1. 5
1.5
22 1. 5
1.5
11
11 0. 50 0
0.5
Node 㢲⛩1 Node 1 㢲⛩5 Node 5 㢲⛩3 3
2.5
࠶Bandwidth 䝽 ᑖ ᇭ (Mbps)
2.5
Bandwidth ࠶ 䝽 ᑖ ᇭ (Mbps)
Mbps 33
Node 㢲⛩1 Node 1 㢲⛩5 Node 5 㢲⛩3 3
2 20 0
4 40 0
6 60
8 80 0Rounds 䖞⅑ 0
10 100
0
12 120
0. 50 0
14 140
0
Fig. 4. Bandwidth allocations under CABMS
P ri c e
0. 0.1 1 0.0 0.09 9 0.0 0.08 8 0.0 0.07 7 0.0 0.06 6 0.0 0.05 5 0.0 0.04 4 0.0 0.03 3 0.0 0.02 2 0.0 0.01 1 0 0 0 0
1 0
0.5
0
2 0
10
3
20
30 Round 0 䖞 ⅑
s
4 0
40
5 0
50
6 0
60
Fig. 5. Bandwidth allocations under proportional fairness
?? 1 1 Clique Clique ? 2 2
2 20 0
4 40 0
6 60
0 Rounds
8 80 0
10 100
0
12 120 0
14 140
0
Fig. 6. Bandwidth prices under CABMS
In figure 4 and figure 5, bandwidth allocation values will convergence after some number of rounds. The bigger of bandwidth allocation step size is and the faster of convergence speed. Data flows F1 and F5 in Q1 and Q2 involve the same number of links. In accordance with rules of CABMS, bandwidth allocation of F1 is 1.5 times the bandwidth of F5, and the bandwidth of F1 is no more than 2.25 times the bandwidth of F3. This is because F3 in Q1
Wang Haitao et al. / Procedia Computer Science 107 (2017) 484 – 489
489
only involves two links, in more competitive Q1, F3 pays lower price than the other two data flows to get the same bandwidth. So it can be allocated more bandwidth. Under the proportional fairness criterion due to occupation of less link resoruces of Q1, general service (F3) get more 50% bandwidth allocation than that of the other two data flows, which makes QoS of two data flows with higher service importance degrades. In the bandwidth allocation CABMS gives data flow (F1) with high importanc greater N value, thus obtaining more bandwidth than that of proportional fairness bandwidth allocation criterion. Obviously, CABMS gets better bandwidth allocation fairness. The changes of bandwidth prices under CABMS is shown in figure 6. As can be seen from the figure 6, since the capacity of Q1 is insufficient to fully meet the bandwidth requirements of the data flows, so bandwidth prices are increasing and eventually stabilizing. For Q2 due to initial larger bandwidth requirements of data flows the bandwidth prices start increasing. But with the changes of bandwidth requirements, the bandwidth prices will gradually be more reasonable. Then, bandwidth capacity of Q2 will fully meet bandwidth needs, and the bandwidth price will drop to zero. 4. Conclusions In practical network application environments, particularly under battlefield circumstances, the importance of various services is different. In bandwidth limited Ad hoc network environment, in order to guarantees successful completion of key services, a context-aware bandwidth allocation scheme named CABMS is proposed. CABMS takes advantage of various sensors to collect data as the basis for context reasoning. Bayesian Network is used as a reasoning tool to infer the importance of services. According to the importance degree of data flows competing scarce bandwidth resources more bandwidth will be allocated to the vital service and thus improving the network survivability. However, to simplify the analysis, in CABMS all variables are assumed to be discrete variables, but in reality some variables (such as noise) are continuous. In addition, in CABMS bandwidth allocation, the clique head node needs collecting bandwidth requirements of all data flows to calculate new bandwidth prices, requiring synchronization among data flows. All these problems are more difficult to solve in dynamic Ad hoc networks. Therefore, an asynchronous bandwidth allocation scheme can be considered for future. References 1. FANG Z, BENSAOU B. Fair bandwidth sharing algorithms based on game theory frameworks for wireless ad-hoc networks[C]//INFOCOM 2004: Twenty-third Annual Joint Conference of the IEEE Computer and Communications Societies. Hong Kong: IEEE, 2004: 1284-1295. 2. XUE Y, LI B, Nahrstedt K. Price-based Resource Allocation in wireless ad hoc networks[C]. Eleventh International Workshop on Quality of Service, Monterey, CA, June, 2003:79-96. 3. CURESCU C, NADJM-TEHRANI S. A Bidding Algorithm for Optimized Utility-Based Resource Allocation in Ad hoc Networks[J]. Mobile Computing IEEE Transactions on, 2008, 7(12):1397-1414. 4. KAO B, LEE L, K. ROBERT LAI. Multi-hop Auction-Based Bandwidth Allocation in Wireless Ad hoc Networks[C] //IEEE International Conference on Advanced Information Networking & Applications. Singapore: IEEE, 2011:772-778. 5. Zhao H, Garcia-Palacios E, Wei J, et al. Distributed resource management and admission control in wireless ad hoc networks: a practical approach[J]. Iet Communications, 2012, 6(8):883-888. 6. DEY AK. Understanding and using context[J]. Personal and ubiquitous computing, 2001, 5(1): 4-7. 7. PERERA C, ZASLAVSKY A, CHRISTEN P, et al. Context aware computing for the internet of things: A survey[J]. Communications Surveys & Tutorials, IEEE, 2014, 16(1): 414-454. 8. WITZIG T, ZOLLNER J M, PANGERCIC D, et al. Context aware shared autonomy for robotic manipulation tasks[C]//Intelligent Robots and Systems (IROS), 2013 IEEE/RSJ International Conference on. Tokyo: IEEE, 2013: 5686-5693. 9. KO K E, SIM K B. Development of context aware system based on Bayesian network driven context reasoning method and ontology context modeling[C]//Control, Automation and Systems, 2008. ICCAS 2008. International Conference on. Seoul: IEEE, 2008: 2309-2313. 10. RHO W H, CHO S B. Context-aware smartphone application category recommender system with modularized Bayesian networks[C]// IEEE Natural Computation (ICNC), Foz do Iguaçu, 2014: 775-779. 11. Anthony D. Wood, John A. Stankovic, Gilles Virone. Context-Aware Wireless Sensor Networks for Assisted Living and Residential Monitoring[J], IEEE Network, 2008,22(4):26-33.