A RELIABLE SEMI-DISTRIBUTED LOAD BALANCING ARCHITECTURE OF HETEROGENEOUS WIRELESS NETWORKS

Transcription

1 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 A RELIABLE SEMI-DISTRIBUTED LOAD BALANCING ARCHITECTURE OF HETEROGENEOUS WIRELESS NETWORKS M. Golam Rabul Alam, Chayan Bswas, Naushn Nower, Mohamme Shaful Alam Khan Islamc Unversty Chttagong, Chttagong-403, Banglaesh gra970@yahoo.com Insttute of Informaton Technology, Unversty of Dhaka, Dhaka-000 Banglaesh chayan_ct@yahoo.com, naushn@t.u.ac.b, shaful@unvhaka.eu ABSTRACT Now a ay s Heterogeneous wreless network s a promsng fel of research nterest. Varous challenges exst n ths hybr combnaton lke loa balancng, resource management an so on. In ths paper we ntrouce a relable loa balancng archtecture for heterogeneous wreless communcatons to ensure certan level of qualty of servce. To conquer the problem of centralze an strbute esgn, a semstrbute loa balancng archtecture for multple access networks s ntrouce. In ths gr base esgn multple Loa an Moble Agent Management Unts s ncorporate. To prove the compactness of the esgn, ntegrate relablty, sgnallng overhea an total processng tme s calculate. An fnally smulaton result shows that overall system performance s mprove by enhancng relablty, reucng sgnallng overhea an processng tme. KEYWORDS Loa Balancng Archtecture, Heterogeneous network, Moble agent, Relablty, Bulletn Boar. INTRODUCTION Toay's wreless networks are hghly heterogeneous wth ntegraton of multple wreless network access nterfaces. For the sake of anywhere-an-anytme connectvty n the next-generaton networks, fferent wreless access networks are ntegrate to form a heterogeneous network []. Wth the growng nterest n varous wreless networks, the nee for ynamc resource utlzaton an effcent transmsson of heterogeneous wreless networks s becomng ncreasngly mportant. Recent research focuses a varety of ssues on heterogeneous wreless networks. Resource management, loa balancng, relablty an QoS support are the most sgnfcant research ssues on what we know as a heterogeneous system. Loa balancng s one of the key technologes n the convergence of heterogeneous wreless networks. It s a sgnfcant way to acheve the resource sharng over heterogeneous wreless networks, an t mproves resource utlzaton, enlargement of system capacty as well as provson better servces for users []. Generally loa balancng epens on network archtecture an the algorthm of loa balancng. In ths paper we prove effcent network archtecture eas DOI : 0.5/jcnc

2 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 so that the loa of an entry system can be properly balance. Network archtecture can be classfe as centralze, strbute, sem centralze an sem strbute [3], [4]. There are some problems n the frst two mechansms: the centralze one s relatvely less relable, whle the strbute one has a large overhea [5]. To overcome the lmtatons of network archtecture we have propose a Sem-Dstrbute Archtecture (SDA) for heterogeneous networks that e ffcently reuces sgnal overhea, mproves relablty an at the same tme utlzes the system loa by ntroucng a moble agent. Moble agents prove a support for ynamc loa balancng snce they can move across heterogeneous platforms an they carry applcaton-specfc coe wth them, rather than requrng pre-nstallaton of that coe on the estnaton machne [6]. Moreover, the agent-base paragm also supports the sruptve nature of wreless lnks an allevates ts assocate banwth lmtatons. The rest of the paper s organze as follows, Secton II revews some of the relate work foun n the lterature. The propose esgn s elaborately epcte on Secton III. The esgn s evaluate n Secton IV. In Secton V smulaton result s llustrate. Fnally, we conclue ths paper n Secton VI wth a scusson.. LITERATURE REVIEW Lots of research was one regarng loa balancng, especally n the area of wre homogeneous networks. But loa balancng n wreless networks s the research ssue of promnence as wreless communcatons are expane ay by ay an multple wreless networks work on the same gr. In the reference [7], a ynamc loa balancng archtecture s propose for heterogeneous wreless network envronments. The authors of the paper have ntrouce the communty concept, n whch rao resources of varous Rao Access Technologes (RATs) are manage together. They have also ntrouce the Communty Resource Management (CRM) functon, the Local Resource Management (LRM) functon, an the Spectrum Manager (SM) n ther archtecture, where CRM buls an sens Communty Resource Announcements (CRA) to Communty Access Ponts (CAP). Here CRM enhances the utlzaton of rao resources an also reuces call blockng probablty of nvual RATs n the communty. The man rawback of ths metho s the complexty of scoverng sutable RATs. Dscoverng a RAT wthout knowng ts operatng nformaton s a tme-an-power-consumng task for Moble Noes (MNs). In ths archtecture, the performance mprovement comes from a vertcal hanoff among RATs, but the cost of a vertcal hanoff s not stue. There s no backup- or falure-hanlng methoology exstng n ths archtecture. In reference [4], a sem-centralze-an-sem-strbute archtecture (SCSDA) s propose. In t, each base staton of any access network type transfers loa nformaton to ts sx neghborng base statons. The archtecture proves mprovements by reucng blockng probablty, roppng probablty an thus avances overall system utlzaton. The major rawback of ths archtecture s the cost of the total number of hanoffs. Both the peroc an non-peroc sgnalng overhea s huge. Moble agent-base loa balancng archtecture s ntrouce n reference [8]. A cret value s assgne wth each of the agents. The assocate cret epens on ts affnty to a machne, current workloa, communcaton behavor an moblty. The cret value of all agents are etermne an then compare to balance loas an f any mbalance s foun then an agent wth 98

3 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 a lower cret value s mgrate to a relatvely lghtly loae machne n the system. As a moble agent s a program whch can autonomously mgrate from one host to another n a heterogeneous network, an can nteract wth other agents or resources [9], we have ncorporate moble agents to balance loas wthn a heterogeneous cell. The authors of [0] consere general heterogeneous network archtecture wth two basc enttes n the system, one s Moble Noes (MNs) an another one s Access Ponts (APs). They formulate the overhea of AP scovery whch s ve nto hello messages an RREQ messages, an gves the smulaton results. The propose archtecture has lack of relablty reflecton. In the paper [], authors ntrouce Herarchcal Sem-Centralze Archtecture (HSCA) for heterogeneous wreless networks. In ths esgn, an Informaton Server (IS), a centralze server, exsts at the top level of the herarchy. An at the mle an lower levels of the herarchy, RA (Resource Allotter) an RS (Resource Statstcs) are ntrouce respectvely, to manage the system loa. The IS s responsble for mantanng backup of all RAs an also accountable for falure-hanlng of RAs. Snce the IS mantans all nformaton of RAs n the system, t creates a huge message-passng overhea. Agan, to fn a proper place for the IS from whch all RAs can access t effcently s qute mpossble. Though there s a backup of the IS, t s not qute effectve to organze a large network wth a sngle IS. 3. DETAILS OF THE PROPOSED DESIGN The etals of the propose archtecture an also the sgnallng flow are scusse n the followng two subsectons. 3.. Propose Archtecture A strbute archtecture for wreless networks s appealng ue to the sngle-pont falure of a centralze esgn. Our esgn s partal/sem-strbute n nature whch s free from the huge overhea an elay problems of a fully strbute esgn. In our propose sem-strbute archtecture, a number of ajacent hexagonal cells create a basc gr. Snce the system s heterogeneous, each cell may consst of fferent access networks. For smplcty, three types of access networks (UMTS, WLAN an WMax) are shown n Fgure. In each gr, a Loa an Moble Agent Management (LMMU) unt s ntrouce. LMMUs consst of three unts; one unt s responsble for mantanng a resource nventory for each type of avalable access network n each cell, whch s calle a Resource Inventory (RI). Another unt, the Moble Agent (MA) [9] n each cell s accountable for balancng loa among varous access networks wthn each cell. 99

4 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 Fgure. Sem strbute archtecture The thr unt of LMMU s Loa an Moble Agent Manager (LMM), whch s responsble for loa balancng among ajacent cells n each gr an among ajacent grs. To ensure relablty, each LMM s assocate wth ts two neghborng LMMs so that f any of the LMMs goes own or s broken, then the backup LMM can take the responsblty for the fale one. In ths propose scheme f one LMM goes own, then only that network management s mantane by a backup LMM an the whole system, therefore, s not hampere. In ths research, we utlze the concept of a Bulletn Boar (BB) [], [3] for mantanng the repostory of Loa nformaton. A Bulletn Boar keeps nformaton for a group of LMMs. To avo a sngle pont of falure, the replca of the BB s mantane, where the number of replca epens on the sze of the network. When an LMM changes ts state t just nforms the BB. To balance loa n the system each LMM exchanges nformaton wth BB. The esgn s calle semstrbute because the loa strbuton ecson s mae by consultng a group of LMMs through the BB. Therefore ths archtecture s free from a large sgnallng overhea. 3.. Sgnallng Flow Usually RI calculates the loa an resources of ts jurscton cell accorng to the wreless parameters receve from the AP, an then transfers the loa, resources an locaton AP RI LMM & Backup LMM Wreless parameters Loa, resources an locaton nformaton Loa balancng nstructons Loa balancng nformaton Fgure. Sgnallng flow type- 00

5 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 nformaton to the LMM. On the bass of loa balancng algorthm, next the LMM transfers loa balancng nformaton to the RI. Then the RI changes loa balancng nformaton to loa balancng nstructons an transfers the loa balancng nstructons to the AP. The sgnallng flow chart s shown n Fgure. Only when LMM changes ts state (uner loae to balance an so on) t then nforms BB. Agan when a LMM searches for sutable LMM, t then asks BB for the better AP RI LMM & Backup LMM Ajacent LMM Wreless parameters Loa, resources an locaton nformaton Moble agent searche for sutable LMMs an transfer necessary loa nformaton. Loa balancng nstructons Loa balancng nformaton one. Fgure 3. Sgnallng flow type- When an MS s on the borer of basc grs, the moble agent wll allocate resources for the MS by consultng wth the neghbourng LMMs, an the sgnallng flow s shown n Fg 3. When an LMM breaks own, the backup LMM can take over mmeately an the resultng sgnalng flow s shown n Fg 4. 0

6 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 4. EVALUATION OF THE PROPOSED ARCHITECTURE To evaluate the propose archtecture we have calculate the ntegrate relablty, sgnal overhea an accumulate tme. Snce these are sgnfcant parameters for any esgn, we have shown comprehensve analyss on them. 4.. Sgnallng Overhea of the Propose Archtecture Sgnallng overhea can be ve as peroc an non-peroc []. Peroc sgnallng overhea s use to backup the nformaton. An non-peroc sgnallng overhea proves loa nformaton to the BB. Here we have stue both peroc an non-peroc sgnallng overhea. 4.. Peroc Sgnallng Overhea For our analyss we use the followng notatons. T Tme of transferrng loa among LMMUs. a The sgnallng overhea of transferrng loa from the RI locate n the system type to the LMM. A The number of APs n system type. Sgnalng overhea of transferrng loa nformaton between the man LMM an the backup LMM. Here, each of the sgnalng overheas s etermne n unt tme. The sgnalng overhea of RIs communcatng to LMMs s: 3 T O a. A The sgnalng overhea of the man LMM communcatng to the backup LMMs s: O * T So, the total sgnalng overhea of transferrng loa nformaton between LMMs s: Let, a a a 3 then 4.. Non-Peroc Sgnallng Overhea O O P 3 O + O + a. A T P * 3 a A + * T Loa nformaton transfers on the BB (whch store loa nformaton for a group of LMMs an where BB reses n one of the LMM) are accorng to the followng rules: a) When the loa state of the LMM changes then the LMM transfers loa nformaton to the BB. From the BB, the LMM can fn loa nformaton of other LMMs. b) When the man BB receves loa nformaton from the LMM then the BB mmeately transfers loa nformaton to backup BBs. () () (3) (4) 0

7 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 Measurement of loa nformaton epens on two threshol values k an k, where k s the lower threshol an k s the upper threshol. If loa of any cell s less than or equal to k then the cell s consere as beng n an uner loae state. If the loa of any cell s greater than or equal to k then the cell s consere as beng n an over loae state. If the loa of any cell s greater than k an less than k then the cell s consere as beng n a balance state. Here we assume that the arrval an servce processes are Posson process an the queung strbuton s M/M/m. So, each cell s an nepenent M/M/m (m) queue. Accorng to the queung theory [4], the state probabltes of system type s: m ( / ) 0 j j P (5) j 0 j! k ( / ) Pk P 0 (6) k! At tme T, the probablty of a cell n system type changng from an uner-loae state to a balance state s: Pr ( U B) Pk T[ ( k ) T] (7) In tme T, the probablty that a cell n system type changng from a balance state to an overloae state s: Pr ( B O) Pk T[ ( k ) T ] (8) In tme T, the probablty that a cell n system type wll change from an overloae state to a balance state s: Pr ( O B) Pk ( k ) T ( T ) + + (9) In tme T, the probablty that a cell n system type wll change from a balance state to an uner-loae state s: Pr ( B U ) Pk ( k ) T ( T ) + + (0) In tme T, the probablty that the loa state of a cell changes n system type s: Pr Pr( U B) + Pr( B O) + Pr( O B) + Pr( B U) () In tme T, the probablty that BB j transfers loa nformaton to backup BBs s: 3 A j Pr ( Pr ) () j The sgnallng overhea of transferrng loa nformaton among the BBs s: 3 O NP a. ( A. Pr ) +. Prj (3) T j 4.. Accumulate Processng Tme of the Propose Archtecture To erve a mathematcal moel for accumulate processng tme, the followng ncators are use. T P Total Processng Tme t Tme requres RI to collect an store loa nformaton (For smplcty, t consere as ms). RL Dstance between RI an LMM LL Dstance between LMM an backup LMM 03

8 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 S RL Propagaton spee between RI an LMM S LL Propagaton spee between LMM an backup LMM W LMM Queung elay at LMM T LMM Average processng elay at LMM Server Utlzaton Factor Average servce tme at LMM Propagaton elay between RI an LMM RL t (4) s Average watng tme at LMM s: W LMM ( ) Average processng elay at LMM s: RL T LMM (6) ( ) So, total queung elay an processng elay at LMM s: t 3 W LMM + T LMM (7) Propagaton elay between LMM an backup LMM s: LL t 4 (8) s W BLMM Queung elay at backup LMM. T BLMM Average processng elay at backup LMM. Average watng tme at backup LMM s: W BLMM ( ) LL Average processng elay s at backup LMM T BLMM ( ) (0) Total queung elay an processng elay at backup LMM s: t W BLMM + T () 5 BLMM W BLMM Queung elay at backup LMM. T BLMM Average processng elay at backup LMM. Average watng tme at backup LMM s W BLMM () ( ) Average processng elay at backup LMM s: T BLMM (3) ( ) So, total queung elay an processng elay at backup LMM s: t W BLMM + T (4) 6 BLMM Total Processng Tme s T P t + t + t3 + t 4 + t5 + t6 (5) (9) 04

9 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 t + S RL RL ( ) ( ) S LL LL ( ) ( ) ( ) ( ) + RL t S RL ( ) S LL LL (5) 4.3. Integrate Relablty of the Propose Archtecture When the number of RIs s large, the breakown of an LMM or the juncton lne between an RI an the LMM has lttle nfluence on the total traffc of the system. Thus the breakown of an RI or the juncton lne between an RI an the LMM can be neglecte. To erve a mathematcal moel for Integrate Relablty, the followng ncators are use. R LMM The relablty of LMM. R C The relablty of juncton lnes between the base LMM an the backup LMM. c j The traffc ntensty between LMMs b j The traffc ntensty between RI an LMM j. n The total number of LMMs. n' The maxmum number of juncton lnes n/3 + n %3 + The probablty that the LMM s n normal use s as follows, conserng base LMM an t s two backup LMMs n p ( ) (6) LMM R LMM Conserng two juncton lnes between LMM j (j,, 3 n') an the two backup LMMs, the probablty that the juncton lne s out of fault s, P (7) C ( R C ) The probablty that the whole system s out of fault s n ' n P P. P. R (8) 0 In that case, the traffc of the whole system oes not lose, (9) L 0 0 The probablty that (K,, 3 n'), juncton lnes between LMMs are out of acton s, n'! P P. R LMM C LMM K n' K LMM. ( PC ). PC K!( n' K)! In that case, the traffc on the K juncton lnes loses, whle all the traffc between LMM s an RI s oes not lose. Thus the whole system traffc loss s, n LMM (30) L K c j j (3) The probablty that K (,, 3 n), whch emonstrates that LMM s are out of acton, s, ' n n! K n k P (3) ( PLMM). Pc ( RLMM). RLMM K!( n K )! Then, there s the nee to calculate the value of another two factors: how the traffc between all 05

10 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 the LMMs nclung losses rrespectve of juncton lnes beng out of acton or not, an the lkelhoo of traffc between the K broken LMM s turnng nto losses. Thus the loss of the whole system s, The total traffc ntensty of the system s L The ntegrate relablty, B n' j n' k j c j + b n j j R [ P0. L 0 + P. L + P. L ] B 3 3 c j + b j j (33) (34) (35) 5. SIMULATION RESULT Peroc sgnallng overhea, non-peroc sgnallng overhea, processng tme an ntegrate relablty are smulate for stuy an comparson purpose. The MATLAB smulator s use to compare the propose SDA wth that of the extng HSCA [], whch s the latest flourshng archtecture of heterogeneous wreless networks over SCSDA [4]. As the HSCA proves better sgnallng overhea an relablty than SCSDA, we conser the HSCA as our benchmark for ths comparatve stuy. 5.. Smulaton Setup an Scenaro For a smulaton purpose an for smplcty, we conser that all three types of APs (UMTS, WMax an WLAN) are present n each basc gr, an the number of APs of smlar network types s equal n each basc gr. It s assume that the network structures are fully overlaye. Intally subscrbers or moble noes are ranomly strbute over all of the three types of wreless networks an we conser ther recton of movement arbtrarly. The followng parameters wth assocate values are use n our smulaton stuy. The number of AP s n UMTS access network (A ) s 600, the number of AP s n WMax access network (A ) s 900 an the number of AP s n WLAN access network (A 3 ) s 600. The tme of transferrng loa among LMMUs s T0.s. We conser, the cell capacty of UMTS cellular system (m ) s 60, WMax system (m ) s 0 an the cell capacty of WLAN system (m 3 ) s 80. To etermne the ntegrate relablty, ntally we conser only juncton lne among LMMs (K ) are out of acton an only LMMs (K ) are out acton. For smulaton stuy, we assume the relablty of each LMM (R LMM ) s 0.9, the relablty of juncton lnes between the base LMM an the backup LMM (R C ) s To smulate the sgnalng overhea of the propose archtecture, we ntally conser the sgnalng overhea of transferrng loa from the RI (locate n the UMTS system) to the LMM (a ) s. Here, we also ntally conser smlar sgnalng overhea of transferrng loa from the RI to the LMM n other two cellular system ( WMax an WLAN) that s a a a 3. The sgnalng overhea of transferrng loa nformaton between the man LMM an the backup LMM () s consere as. 06

11 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 To smulate accumulate processng tme, we conser 0.005ms as the requre tme (t ) of RI to collect an store loa nformaton. For smplcty, both the stance from RI to LMM ( RL ) an the stance from LMM to backup LMM ( LL ) are consere as 500 meters. For smulaton purpose, the propagaton spee between RI an LMM ( S RL ) s assume as 0.5Mbps, whch s equvalent to the propagaton spee between LMM an backup LMM (S LL ). 5.. Smulaton Stuy The smulaton result of the peroc sgnallng overhea s shown n Fgure 5. Here we conser screet number of LMMs as (, 50, 00, 50, 00, 50 an 300). We can see that the peroc sgnallng overhea of our propose system (SDA) remans constant, whereas the peroc sgnallng overhea of the HSCA ncreases wth the ncrement of RAs, an the overhea of HSCA s always hgher than the overhea of the SDA. We can see n [4] that the peroc sgnallng overhea of the SDA s not ncrease wth the ncrease of LMMs because one LMM s connectng only an at most two neghbourng backup LMMs for any network sze. On the other han by ncreasng RA n HSCA, the overhea always ncreases..45 x 04 Peroc Sgnalng Overhea.4.35 Exstng HSCA Propose SDA.3 Sgnalng overhea Number of LMMs or RAs Fgure 5. Peroc sgnallng overhea The sgnallng overhea for a non-peroc sgnal s shown n Fgure 6. Here, the non-peroc sgnallng overhea remans fxe n the propose SDA wth the ncrement of LMMs. On the other han, the non-peroc sgnallng overhea of the HSCA s ncrease exponentally wth the ncrement of RAs. As there s no centralze LMM n the propose SDA, ncrement of LMMs on t have any mpact on the overall non-peroc sgnallng overhea, whch we have analyze n [3]. 07

12 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January Non Peroc Sgnalng Overhea Exstng HSCA Propose SDA 9000 Sgnalng overhea Number of LMMs or RAs Fgure 6. Non-peroc sgnallng overhea Fgure 7 shows the overall processng tme of the propose SDA archtecture wth that of the exstng HSCA archtecture. The processng tme of the HSCA accorng to [36] s much greater than the processng tme of the SDA. In both moels, the processng tme ncreases lnearly wth the ncrement of the traffc arrval rate. The processng tme fference between the HSCA an the SDA also ncreases n lnear fashon Processng Tme Exstng HSCA Propose SDA Processng tme Traffc arrval rate Fgure 7. Overall processng tme Fgure 8 shows the ntegrate relablty of the propose archtecture (SDA) an the ntegrate relablty of exstng HSCA archtecture. The ntegrate relablty s ncrease wth the ncrement of the number of LMMs (RAs) n both of the archtecture, but the ncrement rate of SDA s more faster than the ncrement rate of HSCA. 08

13 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0.00 Integrate Relablty Exstng HSCA Propose SDA Relablty The number of LMMs or RAs Fgure 8. Integrate relablty 6. CONCLUSIONS The propose archtecture s sem-strbute n nature. We have frst propose SDA archtecture n [5], an later, etals of the esgn an assessment of varous parameters are artculate here. From a smulaton stuy, t can be easly known that the SDA performs better by ncreasng relablty an mnmzng overhea an tme. In herarchal sem-centralze archtecture [] there exst only two ISs for the whole network. If they fal to perform then the total network operaton wll be hampere. But the SDA s free from one sngle pont of falure. An t works better for two challengng parameters of strbute system. Obvously, the propose esgn ntrouces an effcent archtecture n heterogeneous wreless network. APPENDIX The authors of the reference [] presente a loa balancng archtecture name HSCA, where they ntrouce a resource management unt (RMU), whch conssts of Resource Statstcs (RS), a Resource Allotter (RA), an Informaton Server (IS) an a backup IS. The RS calculates the loa an resources of ts jurscton cell, the executon of whch requres a certan pero of tme (t ). Then the RS transfers the loa resources an locaton nformaton to the RA, whch requres propagaton elay between the RS to the RA, queung elay an processng elay wthn RAs. Next, the RA transfers the loa resources an locaton nformaton to the IS, whch requres propagaton elay between the RA to the IS, queung elay an processng elay wthn ISs. Smlarly the propagaton elay, queung elay an processng elay wthn the backup ISs exsts when the man IS transfers loa an resource nformaton to the backup IS. Thus the total processng tme s: T + ( ( ) + ) + ( ( ) RR RIS IBI p t RA IS + SRR SRIS SIBI RA IS ) (36) 09

14 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 REFERENCES [] K. Pamrat, A. Ksentn, J. M. Bonnn an C. Vho, (0) Rao resource management n emergng heterogeneous wreless networks, Computer Communcatons, Elsever, Volume 34, Issue 9, pp [] B. L, W. X. Sh an N. L, (00) A Herarchcal sem-centralze archtecture for loa balancng of heterogeneous wreless networks, Secon Internatonal Conference on Networks Securty Wreless Communcatons an Truste Computng (NSWCTC), vol., pp.8 3, [3] Y. R. Lan an T. Yu, (995) A ynamc central scheuler loa balancng mechansm, IEEE Fourteenth Annual Internatonal Phoenx Conference on Computers an Communcatons, pp [4] G. Q. Nng an G. X. Zhu, (006) Loa balancng base on traffc selecton n heterogeneous overlappng cellular networks, Journal of Chnese Computer Systems, vol.7, no., pp [5] G. Q. Nng, G. X. Zhu, L. X. Peng an X. F. Lu, (005) Loa balancng base on traffc selecton n heterogeneous overlappng cellular networks, The Frst IEEE an IFIP Internatonal Conference n Central Asa on Internet. [6] R. Gray, D. Kotz, S. Nog, D. Rus an G. Cybenko, (000) Moble agents: motvatons an state - of -the art systems, Techncal report PCS-TR [7] J. Ha, J. Y. Km, J. U. Km an S. H. Km, (009) Dynamc Loa balancng Archtecture n Heterogeneous Wreless Network Envronment, n Proc. of 9th Internatonal Symposum on Communcatons an Informaton Technology (ISCIT), Sept. 8-30, South Korea, pp [8] P. K. Snha an S. R Dhore, (00) Mult-agent optmze loa balancng usng spannng tree for moble servces, Internatonal Journal of Computer Applcatons ( ), Volume No. 6, pp [9] W. X. an Z. Q. Qang, (00) The analyss of the applcaton about MA n E-Commerce, Internatonal Conference on E-Busness an E-Governance, pp [0] K. Sharf, L. J. Cao, Y. Wang an T. Dahlberg, (008) A Hybr any cast routng protocol for loa balancng n heterogeneous access networks, Internatonal Conference on Computer Communcatons an Networks, ICCCN 008, pp [] W. Sh, B. L, N. L an C. Xa, (0) A Network archtecture for loa balancng of heterogeneous wreless network, Journal of Networks, Volume 6, No 4, pp [] P. K. Snha, (005) Dstrbute Operatng Systems: Concepts an Desgn, Ffth Eton, Prentce- Hall of Ina, pp [3] I. Rahman, N. Nower et al., (0) Optmal loa balancng of herarchal structure multple home agents, Internatonal Journal of Computer an Informaton Technology, Vol., Issue, pp [4] W.X. Sh, L. C. Zhang, K. G. Hu an Y. Dong, (008) Communcaton network: prncple an applcatons, Publshng House of Electroncs Inustry, Bejng, Chna, pp [5] M. G. R. Alam, C. Bswas an N. Nower, (0) A Sem Dstrbute an Moble Agent Base Archtecture for loa balancng of Heterogeneous Wreless Networks, Internatonal Conference of Computer an Informaton Technology ICCIT 0 (n press). 0

15 Internatonal Journal of Computer Networks & Communcatons (IJCNC) Vol.4, No., January 0 Authors M Golam Rabul Alam receve the Bachelor Degree n Computer Scence & Engneerng from Khulna Unversty n 00 an the Master Degree n Informaton Technology from Unversty of Dhaka n 0. He s now an Assstant Professor of Department of Computer Scence an Engneerng n Internatonal Islamc Unversty Chttagong (IIUC). Hs current research nterests nclue the Heterogeneous Wreless Networks, Data Mnng an Informaton Securty. He s a co-author of about thrty research papers an a text book. Chayan Bswas receve the Master Degree n Informaton Technology from Unversty of Dhaka n 0. He s now a Senor Offcer (IT) of Banglaesh Development Bank Lmte (BDBL). Hs current research nterests nclue the Loa Balancng technology for Wreless Heterogeneous Networks an Wreless Securty.