Load balancing over redundant wireless sensor networks based on diffluent Abstract Xikui Gao Yan ai Yun Ju School of Control and Coputer Engineering North China Electric ower University 02206 China Received August 204 www.cnt.lv As the traffic blocking probability of traditional hard load balancing algoriths is generally high over redundant wireless sensor networks This paper proposes a load balancing algorith based on dividing the packet flow (LD) over redundant wireless sensor networks based on the idea of soft load balancing. In the schee through nuerical analysis obtain the optial flow-dividing ratio to deterine the volue of traffic delivered to each network which aintains network load balance. Fro the theoretical and siulative perspectives the paper analyses the perforance paraeters and the analytical results show that the perforance of the schee is better than other schees. Siulation results show that the proposed ethod outperfors traditional hard load balancing techniques in ters of traffic blocking and packet loss probabilities. Keywords: redundant wireless sensor networks load balancing packet flow diversion access selection handover Introduction The future of counication network is a redundant structure syste various wireless access network access to public core network based on all I through a variety of different access technology []. Redundant wireless sensor network resource anageent technology especially the load balance technology is one of the key technologies to realize integration in the redundant wireless network and is a research hotspot in recent years. Load balancing is the iportant ethod to ipleent effective resource sharing in the redundant wireless network which can iprove the utilization of wireless resources in redundant wireless networks expand the syste capacity to provide users with better service. References [2 3] consider syste load affected by the users the condition of channel the QoS (quality of service) requireents of business and other factors and put forward when the syste load is greater than the prescribed threshold ethod the high load transfer part of the user in the syste to the low load syste. In reference [4] calculate the need of resources consued by users access each network the design target of access is. After the user access network the ratio of consuption of resources and network resources available to is the iniu; reference [5] used ulti-obective decision-aking ethod for network selection in the redundant wireless network. Discuss the load balance and other paraeters of coproise; in a redundant and distributed grid environent reference [6] designed a kind of iune clone task scheduling algorith to achieve the balanced allocation of resource and efficient scheduling tasks. The above references about ethod of load balancing are all hard load balancing - The users business at the sae tie can only access a network not well QoS guarantee business business blocking ratio is higher [7]. References [7 8] propose the concept of soft load balance. The packet of down link can be divided into sub strea each child flows into the different wireless access networks this ethod can ore fully use of resources in the redundant wireless network. ut references [7] only studied the best diversion ratio in specific network topology circustances not universal; and reference [8] only studied the diversion ratio under optial wireless channel environent the results deviation with the actual situation. Aiing at the existing proble of the load balance ethod This article propose the load balancing algorith under the whole cover network environent based on packet flow diversion (load alancing algorith based on dividing packet flow LD). Algorith considered the effect of ultipath and path loss user service in redundant wireless network is obtained by nuerical analysis of the optial split ratio. When users need to access network or switch syste downstrea business will be divided into the subflows according to the optial split ratio and the subflow will access or switch to the corresponding network to iprove the network capacity iprove the syste perforance. Corresponding author e-ail: uyun982@ncepu.edu.cn 490
2 The syste odel 2. SOFT LOAD ALANCE NETWORK STRUCTURE In this paper on the basis of reference [9] iproved a layered half centralized network structure as shown in Figure. Several location adacent area apping into a basic grid inforation server-is resources allocation-ra and resources statistics-rs are collectively referred to as resource anageent unit-rmu it is responsible for the anageent of grid resources. RS is set to access node in the node for statistics and calculation of the resource inforation; RA is used to collect the RS inforation and according to the situation of the node load and basic grid resources for load balancing. IS as RA supervisor server basic grid structure under the noral state only responsible for edges of resource allocation and the identification of the storage nodes the inforation such as position load condition but when the anageent of RA alfunction cant work IS can quickly take over and perfor the function of RA. IS RA RS WiMax UMTS WSN RA Main IS Sub IS asic Grid asic Grid 2 FIGURE Hierarchical sei-centralized architecture in a twodiensional resource unit can get the biggest transfer rate. In order to support the soft load balancing. The public link layer are introduced to processing access control signals in RA [7]. In each of the access network below the public link layer is the edia access control layer and physical layer the MAC layer is located in the RS. Each MAC protocols report it channel state to the public link layer ublic link layer deterine the size of the diversion ratio according to the obtained channel condition and syste load conditions. The public link layer data received fro the MAC layer according to the data of the serial nuber on the reorder and then to transfer the data through the network layer to the transport layer. 2.2 TWO-DIMENSIONAL RESOURCE UNIT Syste odel have M kinds of overlapping coverage of access syste different access syste adopts OFDM (orthogonal frequency division ultiplexing) technology. For syste ( 2 M) all subcarrier in sequential ode can be divided into a nuber of sub-channels. Each channel contains F carrier. The frae length T f is RA2 divided into isoetric tie slot a tie slot contains S OFDM sybols. Such a tie slot on the tie doain and frequency doain on a child channel will constitute a twodiensional resource unit as the basic unit of the resource allocation in OFDM syste [4]. In a two-diensional resource unit can get the biggest transfer rate: F SCi bi () T f where after the user i access syste. ased on adaptive coded odulation echanis to deterine the inforation bits C i carried by each odulation sybols [0]. 2.3 CHANNEL MODEL Considering the influence of ultipath and path loss channel odel of this paper by using the Rayleigh fading odel the user receives the base station signal power [4]: R Ad T 2 (2) where T for the access point transission power. In this paper redundant wireless sensor networks in each syste assuing that access point transission power is constant and evenly distributed in the entire history of the available frequency band; α is the ean of and obey exponential distribution of rando variables used to characterize the fast fading. Ad β is used to represent the path loss constant A decided by the antenna height and carrier frequency β is attenuation index d is the distance between the user and the access point. Assue that the user i in the syste within the coverage of the access point the dry ratio γ i at the receiving end can be expressed as: i N k Ad 2 2 Ad i i T T i (3) where ω i is the noise power at the receiving end of user i in the syste d i is the distance between user i and the access point of syste N is the nuber of other access points which are the disturbance of access point of syste T is the transission power of access points in the syste. In the AWGN channel with M - QAM odulation the biggest transport bit error rate fro the access point to the user i in the syste [4] is: ER i.5 i C 2 i e. (4) 5 When the packet length is K bits in order to obtain the relationship between the packet loss rate and bit error rate considering the liiting cases. Assue that after through other error correction echanis any one bit error eans 49
that the entire packet transission error the packet loss rate of the user i in the syste is []: L i. ER i K. (5) Considering autoatic retransission echanis of MAC layer an average transfer nubers of receiving a right packet group took [4]: K. (6) L i 3 The load balancing algorith based on packet flow diversion When the user have soe group business need network access or in the process of service because of in the syste or between the syste switch due to obile factors such as the decrease of the quality of the business requireents Syste for its first calculated optial split ratio then according to the optial split ratio to access or switch to the appropriate network the algorith overall process is shown in Figure 2 algorith steps as shown in section 2.3. 3. TWO-DIMENSIONAL RESOURCE UNIT ALLOCATION MODEL Assue the rate deand of user i is R i user i business in syste transission rate is R i the diversion ratio of users i in the syste is ω i the Equations (7) and (8) are as follows: R R (7) i i i i (8) After user i access syste each frae required twodiensional resources unit for an average is: R ni ki (9) bi where x is the iniu integer greater than or equal to x. User requests access networks/switch Assue that the user i access syste before the access point the access point is covered inside the village which have accessed I I users the load in the location can be expressed as: L I i n. (0) i k k In order to satisfy the QoS requireents of group business this paper adopts the following strategies: ) The average service rate of the non real-tie business is greater than the iniu service rate on the basis of using the best scheduling echanis. So in access to choose Equation (9) can be said it took an average of network resources. 2) Most of the grouped data real-tie business to send in a axiu tie delay constraint. This paper adopts the effective bandwidth theory to ensure real-tie QoS requireents of the business If the bandwidth of the network to provide real-tie business is greater than the effective bandwidth its QoS guaranteed [2]. Assue that grouping package (on) the activation state only at fixed intervals and in the inactive state (off) does not produce. Suppose syste on his stay on and off states respectively obey negative exponential distribution of paraeters for the λ and μ the probability of packet delay is greater than the given threshold D to ε the user i equivalent bandwidth of real-tie business can be expressed as [3]: Ri ( D ln ) Wi. () ( ) D ln If the network to provide real-tie business available data rate is greater than the type of effective bandwidth value in Equation () the real-tie business QoS can be guaranteed. y Equation (9) real-tie business users i access syste 2d resources required for each frae after the average cell nuber is: n k W. (2) bi 3.2 THE EST DIVERSION RATIO METHOD set up two-diensional resource unit allocation odel The optial split ratio according to the optial split ratio control user access networks/switch This paper takes the real-tie business as an exaple the optial distribution ratio real-tie business best solution of the diversion ratio is siilar. Set up an access point in the syste available unit for a total of 2d resources L T define the user i to the influence degree of the network load after access networks (hereinafter referred to as the load factor) is: n M i. (3) LT. FIGURE 2 Flow chart of LD 492
In order to get the best load balancing effect user i in the syste the best split ratio is obtained by Equation (4): M i.. i M arg ini arg in. (4) LT Load factor expressions η i is received using Equations (4) (5) (7) (9) and (3): n Step 3 when the su of all the available resources in syste l and l cannot eet the deand of user business resources LT l Ll i i l L L and T l l i i l i l siilar to the step 2 calculation values other than = l. Step 4: if the su of all syste resources available cannot eet the deand of users i rate R i declined to the users access to the switch or request. M i i i (5) LT where: R b L i i. (6) When the user i access networks R i Li and b i are constant values so i for the fixed value. est diversion ratio of the calculation steps are as follows: Step : when l arg in i access points of... L T syste l have enough resources to allow users i all access L L get the obective function for when T l l i i l solutions of Equation (4): l arg in... LT. (7) 0 other Step 2 when the access point of syste l does not have enough resources in to allow user i all access when LT l Ll i i l ake ni l LT l Ll i you can get: L T l Ll i i il (8) i l arg in... L ake l T arg in l... L T if the first access points of the syste l have enough resources to the user i full access L L then: when T l l i i l i l 3.3 LOAD ALANCING STES LD algorith steps are as follows: Step : when a user i need access to the network or switch RA or IS (when user i on the verge of basic grid) scanning for the available access points list of users i. Step 2: RS according to Equation (3) to easure Signal to Interference Ratio γ i fro user i to each available access points. According to the Equations (4) and (5) channel bit error rate and packet loss rate are calculated respectively. Cobined with the corresponding link adaptive echanis with Equation () to calculate the corresponding each 2d resource unit can obtain the axiu transission rate of b i. To transfer the result of the easureent and calculation for RA. Step 3: RA through using the ethod of section 2.2 to calculating optiu diversion ratio for the user i. Step 4: RA according to the optial split ratio to divide users i business into sub strea and flow or switch to the corresponding access to the network. 4 Algorith perforance analysis 4. THE SIMULATION ANALYSIS roposed by reference [4] ML (axiu load balancing algorith is hard load balancing ethod of superior perforance of an algorith in order to verify the LD algorith perforance in this paper the LD algorith and the ML business traffic congestion packet loss rate of the algorith and the network load balancing degree has carried on the siulation analysis. The siulation scenario is shown in Figure 3 [4]. LT l Ll i i l arg in i l... LT i i l l arg in l... L 0 other T. (9) WiMax 3G LTE FIGURE 3 Siulation scenario 493
LTE syste area (.5 k) radius and WiMax syste area (radius 3 k) overlapping coverage two systes centre village access point location collocated. At the beginning of the siulation the user rando distribution in the region of the repeat coverage rando direction oved to the village edge handover will happen. When user ove at speed 20 k/h users location and oveent direction is updated once every 00 s one direction change ±45 at 0.2 probability at a tie. Each users business source are independent of each other and with the sae probability to generate real-tie and non real-tie business. Siulation paraeter Settings as shown in Table. access two network at the sae tie so its block probability ML was always lower than the algorith and the ore users the ore obvious advantages in block probability LD algorith. Figure 6 for the user to access the network or when switching the average packet loss rate. It can be seen that with the increase of users the LD algorith proposed in this paper the packet loss rate than ML always low packet loss rate of the algorith. Figure 7 for two networks under two kinds of algorith copares the average noralized load two algoriths is visible in the aspect of load balance ability has been close to two network load conditions. TALE Siulation paraeters settings nae UMTS WIMAX Carrier frequency/bandwidth 2.0 GHz/0MHz 3.5GHz/0MHz T f 0 s 0 s OFDM sybol cycle 0.2 s 0.2 s NFFT 52 024 T 43d 46d W i -70 d/hz -74 d/hz 24 consecutive 24 consecutive Two diensional subcarrier and 0 subcarrier and 0 resource unit sybols of cycle sybols of cycle K 52 bit 52 bit Unifor distribution in [50 kb/s 200 kb/s] Real-tie business rate Non real-tie business rate Unifor distribution in [50 kb/s 200 kb/s] Unifor distribution in [40 kb/s 40 kb/s] Unifor distribution in [40 kb/s 40 kb/s] λ /μ 2.84 s/.53 s 2.84 s/.53 s usiness generate tie interval usiness last tie interval for s for 2 s for s for 2 s nae UMTS WIMAX Carrier frequency/bandwidth 2.0 GHz/0MHz 3.5GHz/0MHz The statistical siulation results statistics only two network repeat coverage areas darker areas in (Figure 3). According to the paraeters of Table Through the Matlab siulation results as shown in Figure 4 ~ 7 users in the figure is a LTE area within the scope of active users. Figure 4 for the nuber of I when using the algorith of LD shunt and the ratio of total access and switching ties. When users less than 30 have enough resources to accept user network so packet flow without shunt. With the increase of nuber of users a single network resources began to cant satisfy user needs of the business user packet of the business flow is divided into two sub flow access two network at the sae tie. Figure 5 for the network load is full cause user blocking probability. Due to network users less than 30 have enough resources to acceptance of users users under two kinds of algorith the block probability is 0. With the increase of users the users block probability LD algorith can be very high in both the network load will users business packet flow is divided into two sub flow diversions ratio Access and handoff blocking probability packet loss probability 2.5 2.5 0.5 0 0 20 30 40 50 60 70 80 90 The nuber of users 0. 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.0 Flow-dividing tie FIGURE 4 Flow-dividing tie. 0 0 20 30 40 50 60 70 80 90 The nuber of users 0.09 0.085 0.08 0.075 0.07 0.065 0.06 0.055 ML LD FIGURE 5 locking probability. 0.05 0 20 30 40 50 60 70 80 90 The nuber of users FIGURE 6 acket loss probability. ML LD 494
Load balancing degree 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 FIGURE 7 Load balancing degree. 4.2 ALGORITHM ALICAILITY Reference [7] study the network structure of two kinds of syste coverage crisscross overlap neighbourhood radius of two systes the sae access point located at the centre of the counity and a syste of counity centre is located in another syste on the edge of the neighbourhood see Figure 3 in reference [7]; In this paper we study the network structure as the centre of counity access point phase coincidence of the two systes and neighbourhood radius of two systes can be different this structure is closer to the actual network. Reference [7] only analyses the two systes centre village access point best References ML() ML(2) LD() LD(2) 0. 0 20 30 40 50 60 70 80 90 The nuber of users in the iddle of the diversion ratio; Algorith in this paper to consider all the best shunting overlapping coverage ratio scope of application is ore widely. Reference [8] only studied the ideal of the wireless channel environent load balance; In this paper considering the influence of ultipath and path loss. Therefore copared with the existing soft load balancing algorith this algorith is ore universal. 5 Conclusions In this paper based on packet flow diversion of heterogeneous wireless network load balancing algorith. Algorith considering the whole network environent the real-tie statistics network status indicators calculated under different network load balancing target users access to the network or the best split ratio when switching. Choose according to optial split ratio for access or switch can achieve less resource consuption iprove the syste capacity. The siulation results show that LD algorith can effectively balance the network load and the group business average blocking probability and packet loss spontaneously in a relatively traditional ML algorith has obvious iproveent. Acknowledgents This work is supported by the Fundaental Research Funds for the Central Universities of China (204QN). [] Javaid N Qureshi T N Khan A H Iqbal A Akhtar E Ishfaq M 203 EDDEEC: Enhanced Developed Distributed Energy-Efficient Clustering for Heterogeneous Wireless Sensor Networks rocedia Coputer Science 9 94-9 [2] Hefeida M S Canli T Khokhar A 203 Cl-ac: A cross-layer ac protocol for heterogeneous wireless sensor networks Ad Hoc Networks () 23-25 [3] Szczodrak M Gnawali O Carloni L 203 Dynaic reconfiguration of wireless sensor networks to support heterogeneous applications IEEE International Conference on Distributed Coputing in Sensor Systes (DCOSS) 203 52-6 [4] Al-Haadi H Chen R 203 Redundancy anageent of ultipath routing for intrusion tolerance in heterogeneous wireless sensor networks IEEE Transactions on Network and Service Manageent 0(2) 89-203 [5] Alonso R S Tapia D I ao J García O de az J F Corchado J M 203 Ipleenting a hardware-ebedded reactive agents platfor based on a service-oriented architecture over heterogeneous wireless sensor networks Ad Hoc Networks () 5-66 [6] Shara K K 203 Traffic contention assessent and control for iproveent of QoS in heterogeneous wireless sensor networks In fulfillent of the requireents for the degree of doctor of philosophy in electronics & counication engineering suary of report [7] Wang X Han S Wu Y Wang X 203 Coverage and energy consuption control in obile heterogeneous wireless sensor networks. IEEE Transactions on Autoatic Control 58(4) 975-988 [8] arvatkar S U Gore D V 204 An Ant Colony Optiization Algorith for axiizing the lifetie of Heterogeneous Wireless Sensor Network [9] Ki D Y Cha S H Cho K H 203 Ontology-based ethodology for anaging heterogeneous wireless sensor networks International Journal of Distributed Sensor Networks 203 60684 [0] Maia G Guidoni D L Viana A C Squino A L L Mini R A F Loureiro A A F 203 A distributed data storage protocol for heterogeneous wireless sensor networks with obile sinks Ad Hoc Networks (5) 588-602 [] Kapoor N K Nandy Maudar S 203 Dynaic allocation of sensor nodes in wireless sensor networks hosting ultiple applications IEEE International Syposiu on erforance Evaluation of Coputer and Telecounication Systes (SECTS) 203 5-7 [2] Senturk I F Akkaya K Yilaz S 204 Relay placeent for restoring connectivity in partitioned wireless sensor networks under liited inforation Ad Hoc Networks 3 487-503 [3] Younis M Senturk I F Akkaya K Lee S Senel F 204 Topology anageent techniques for tolerating node failures in wireless sensor networks: A survey Coputer Networks 58 254-83 [4] Tapia D I Alonso R S García Ó de la rieta F érez-lancho 203 Cloud-IO: Cloud coputing platfor for the fast deployent of services over wireless sensor networks The 7th International Conference on Knowledge Manageent in Organizations: Service and Cloud Coputing Springer erlin Heidelberg 72 493-504 [5] Chen D Liu Z Wang L Dou M Chen J Li H 203 Natural disaster onitoring with wireless sensor networks: a case study of dataintensive applications upon low-cost scalable systes Mobile Networks and Applications 8(5) 65-63 495
Authors Xikui Gao born in October 959 China Current position grades: senior engineer professor. University studies: School of Control and Coputer Engineering North China Electric ower University. Scientific interests: control theory and control engineering. Experience: resident of haronious technology co. LTD. Jiangsu China. Yan ai born in March 954 Shenyang Liaoning province China Current position grades: Doctoral degree professor at the School of Control and Coputer Engineering North China Electric ower University. University studies: northeastern university hd. Scientific interests: power plant production process autoation design and teaching. ublications: ore than 50. Yun Ju born in August 982 Lichuan Jiangxi province China Current position grades: doctoral degree lecturer at the School of Control and Coputer Engineering North China Electric ower University. University studies: hd at the School of Control and Coputer Engineering North China Electric ower University. Scientific interest: control theory and control engineering. ublications: ore than 7. 496