Control of Wireless Networks with Flow Level Dynamics under Constant Time Scheduling


 Lee Cunningham
 1 years ago
 Views:
Transcription
1 Control of Wirele Network with Flow Level Dynamic under Contant Time Scheduling Long Le and Ravi R. Mazumdar Department of Electrical and Computer Engineering Univerity of Waterloo,Waterloo, ON, Canada N2L 3G Abtract We conider the network control problem for wirele network with flow level dynamic under the general k hop interference model. In particular, we invetigate the control problem in low load and high load regime. In the low load regime, we how that the network can be tabilized by a regulated maximal cheduling policy conidering flow level dynamic if the offered load atifie a contraining bound condition. Becaue maximal matching i a general cheduling rule whoe implementation i not pecified, we propoe a contanttime and ditributed cheduling algorithm for a general khop interference model which can approximate the maximal cheduling policy within an arbitrarily mall error. Under the tability condition, we how how to calculate tranmiion rate for different uer clae uch that the longterm time average) network utility i maximized. Thi longterm network utility would capture the real network performance due to the fact that under flow level dynamic, the number of uer randomly change o intantaneou network utility maximization doe not reult in ueful network performance. Our reult imply that congetion control i unneceary when the offered load i low and optimal uer rate can be determined to maximize uer longterm atifaction. In the high load regime where the network can be untable under the regulated maximal cheduling policy, we propoe the crolayer congetion control and cheduling algorithm which can tabilize the network under arbitrary network load. Through extenive numerical analyi for ome typical network, we how that the propoed cheduling algorithm ha much lower overhead than other exiting queuelengthbaed contanttime cheduling cheme in the literature, and it achieve performance much better than the guaranteed bound. In addition, uing congetion control in the low load condition reult in much lower average utility compared to that due to the optimal tranmiion rate derived in the paper. Index Term Flow level dynamic, capacity region, contanttime cheduling, network tability, maximal matching, khop interference model I. INTRODUCTION Reource allocation in communication network ha been an active reearch topic for the lat everal year. While optimal rate control in wired network can be achieved by a ditributed algorithm [][4], olving thi problem in wirele network i much more challenging. In fact, the bottleneck of the reource allocation problem in wirele network lie in the cheduling ubproblem [5][6]. The difficulty of the cheduling ubproblem come from the interference coupling of imultaneou tranmiion from different wirele link in the network. In general, interference coupling in wirele network depend on the communication technologie employed at the phyical layer. For example, the node excluive interference model can be aumed for Bluetooth network or FHCDMA network [7], [8]. Thi interference model i alo referred to a onehop interference model. Alo, for the 802. WLAN with fourway handhake i.e., with RTS/CTS), the twohop interference model i implicitly aumed in the MAC protocol. Moreover, it ha been hown in [9] that in certain network etting and QoS requirement, neither onehop nor twohop interference model i optimal to achieve the maximum number of imultaneou tranmiion in the network. In [9] the author propoed a general interference model called khop interference model which i determined by a ingle parameter k. For thi interference model, wirele link k + or more hop away from one another can be cheduled to tranmit data at the ame time. Developing a joint reource allocation and cheduling algorithm for thi khop interference model i, therefore, much more deirable than working with a pecial cae of thi general interference model. Thi i indeed what we will purue in thi paper. Regarding the cheduling problem in wirele network, there are everal optimal and uboptimal cheme propoed in the literature. In a eminal paper [0], Taiula and Ephremide propoed an optimal backpreure policy which achieve the maximum network throughput. Thi cheduling policy i, however, centralized and computationally expenive. In [], a randomized linearcomplexity cheduling algorithm wa propoed where a tranmiion chedule in time lot t wa contructed by chooing the chedule with larger total weight between the chedule in time lot t and a newlygenerated one in time lot t. Thi idea wa ued to develop ditributed throughputoptimal cheduling policie in [2] [4] for onehop and twohop interference model. Note that thee cheduling algorithm achieve full utilization of wirele network with repect to what remain in the data tranmiion phae only. Specifically, a large amount of bandwidth ha been wated to exchange control information in the chedule contruction phae which would otherwie be ued for data tranmiion. In general, the amount of cheduling overhead grow with the network ize for thee throughputoptimal cheduling policie. Due to implementation contraint, the time lot interval i uually limited to a few milliecond a in mot current wirele ytem. Therefore, developing a cheduling algorithm with low and contanttime overhead would be very deirable. In fact, ome queuelengthbaed contanttime cheduling algorithm were propoed for one and two hop interference
2 2 model [5][7] recently in the literature. Thee cheduling algorithm only achieve a guaranteed faction of the capacity region but they have contant time overhead. For practical implementation, collecting queue length information may be difficult, and it will create further overhead. A more general maximal cheduling policy wa conidered in [9], [20] where everal throughput performance bound were invetigated. However, implementation of thi general cheduling policy and invetigation of it actual performance in typical wirele network were not conducted in thee paper. In practice, it i deired that each wirele node only communicate with it neighbor e.g., thoe whoe tranmiion interfere with that of the underlying node) to contruct a tranmiion chedule in each time lot. Alo, cheduling algorithm hould work for a general cla of interference model e.g., khop interference model [9]). Another apect which wa ignored by mot exiting work in the literature i that no conflictfree chedule i available to exchange control information at the beginning of each time lot. Therefore, control information can only be exchanged by uing contentionbaed tranmiion which render information exchange more than one hop away a timeconuming operation. Alo, it i important to quantify amount of time/overhead ued to contruct the chedule and to develop explicit procedure to exchange control information in each time lot. In thi paper, we how the performance guarantee of the regulated maximal cheduling policy in wirele network conidering flow level dynamic. In fact, regulated maximal cheduling i a combination of the maximal cheduling policy [9], [20] and a traffic regulator implemented at each wirele link. Becaue regulated maximal cheduling i a general rule, we propoe a contanttime and ditributed algorithm to implement it in each time lot. We how that the propoed cheduling algorithm can approximate the regulated maximal cheduling policy within an arbitrarily mall error. The propoed cheduling algorithm work for a general k hop interference model and doe not require queue length information. Moreover, we explicitly decribe how wirele link coordinate their contention to contruct the chedule in each time lot. The aforementioned performance guarantee reult implie that we do not need to perform congetion control even with flow level dynamic if the traffic load lie within a region which can be tabilized by the underlying cheduling algorithm. Thi i an intereting finding given the fact that there are exiting work which employ congetion control algorithm to tabilize the network [5], [2] under low network load. Given the fact that the cheduling algorithm i a randomized one and the number of uer dynamically change, intantaneou network utility maximization would not reult in good network performance. Becaue of thi, we are intereted in finding tranmiion rate for each uer cla which achieve maximum longterm time average) utility. In fact, we how that there exit optimal tranmiion rate for all uer clae to achieve uch maximum longterm utility. When the network load i high, we propoe a crolayer congetion control algorithm which can tabilize the network for arbitrary network load. The reult preented in thi paper have everal important implication for ytem implementation. Firt, we do not need to perform congetion control in low network load even with flow level dynamic. Hence, communication overhead a well a implementation iue uch a aynchronou [8] and noiy [24] feedback due to congetion control operation can be completely avoided. Second, the problem of network utility maximization can be decoupled from that of tabilizing the network. Specifically, the network can be tabilized by implementing traffic regulator at wirele link together with a uitable cheduling mechanim. In fact, we how via numerical example that uing congetion control algorithm to tabilize the network in low load actually degrade the longterm network utility coniderably. The remaining of thi paper i organized a follow. We decribe the ytem model and performance bound in ection II. Performance guarantee of the regulated maximal cheduling policy i preented in ection III. In ection IV, we preent the ditributed cheduling algorithm to approximate the maximal cheduling policy. We derive the optimal tranmiion rate to achieve longterm utility maximization in ection V. The crolayer congetion control and cheduling algorithm in the high load regime i decribed in ection VI. Some numerical reult are preented in ection VII and ection VIII tate the concluion. For notational convenience, we will put element of different meaure into the correponding vector. For example, the vector of tranmiion rate will be denoted by x where x i it th element which i the tranmiion rate of cla uer. II. SYSTEM MODELS AND PERFORMANCE BOUND We model a wirele network a a directed graph G = V, E) where V i the et of wirele node and E i the et of wirele link. A wirele link from node i to node j exit if node j can correctly receive information tranmitted by node i. In practice, exitence of uch a link depend on tranmiion power, path lo, fading, interference, deired bit error rate and other factor. We aume that there are S clae of uer each of which aociate with a fixed routing path from a ource node to a detination node. The uer route are tored in an incidence matrix [H k ] where H k = if link k i on the route of cla uer and H k = 0 otherwie. Uer of cla arrive to the network with rate λ and each bring a file for tranfer whoe ize i exponentially ditributed with mean /µ. The offered load by cla uer i, therefore, ρ = λ /µ. The vector of offered load will be denoted a ρ = [ρ, ρ 2,, ρ S ]. We aume that uer of each cla tranmit at the ame rate. Interference contraint are denoted by a contention matrix [C ij ] i,j E. Specifically, link i i aid to interfere with link j if C ij = and C ij = 0 otherwie. Thi general notation of the interference relationhip i ued to decribe the capacity region and to derive the performance guarantee of the regulated maximal matching policy in thi ection and ection III only. The khop interference model i a pecial cla repreenting thi general interference relationhip. The khop interference model i aumed in all remaining ection of thi paper. Time i divided into lot of unit duration. Link l can tranmit at
3 3 rate if it interfering link are not cheduled to tranmit in a ame time lot. In the following, we provide ome important definition which will be ued in the equel. Definition I: Interference et I l of link l i the et of link which interfere with link l, i.e., I l = {k E : C kl = }. ) Definition II: Interference degree d I l) of link l i the maximum number of link in it interference et which do not interfere with each other. Definition III: Interference degree d I G) of graph G i the maximum interference degree of it link, i.e., d I G) = max l E d I l). Capacity region i defined to be the et of traffic load uch that the network can be tabilized by ome cheduling policy. In [0], capacity region for wirele network i well characterized. In particular, capacity region i given by the et { [ S ] } H l Ω = ρ : ρ CoR) 2) R l l E where CoR) i the convex hull of all link chedule R that atify Wirele the linkregulator buffer RegulatorTranmiion queue contraint impoed by the underlying interference Wirele link model. A cheduling policy i aid be throughput optimal if it tabilize the network for all offered load within the capacity region Ω. Fig.. Regulator implementation at each wirele link. We aume that a new chedule i contructed in the firt phae of each time lot which i ued to tranmit data in the econd phae of each time lot. Traffic flow in the network may travere one hop or multiple hop. For the multiplehop cae, we aume that traffic of each uer cla i regulated before entering a tranmiion queue for being tranfered over each wirele link. The employment of regulator wa previouly propoed by Hume to tabilize manufacturing ytem [22] which have been hown to be untable in ome cae due to cycle of material flow [23]. Regulator were recently ued in wirele network [8], [9]. A λregulator aociated with link l generate packet to tranmiion queue of link l with a maximum rate of λ. A regulator can be implemented a follow. In each time lot, a λregulator aociated with link l check the correponding regulator queue. If the queue length i greater than link capacity then it tranfer unit of data to the tranmiion queue with probability λ/. Otherwie, it tranfer nothing. The regulator implementation i illutrated in Fig.. In thi paper, we aume a maximal cheduling policy which wa invetigated in [9], [20]. The maximal cheduling rule can be decribed a follow. For any link l E with tranmiion queue length larger than the link capacity in any time lot, it i required that at leat one link in it interference et I l be cheduled. Specifically, if Q l / where Q l i the queue length of tranmiion queue for link l), we require π k 3) where π k = if link k i cheduled and π k = 0 otherwie. Due to the combination of maximal cheduling and regulator implementation, the reulting cheduling will be called regulated maximal cheduling in the equel. Note that maximal cheduling i a general cheduling rule without pecific implementation. We will preent a contant time and ditributed cheduling algorithm which approximate the maximal cheduling in ection IV. The following performance bound of the maximal cheduling policy wa proved in [20], [2], it i retated here for completene. Lemma : For all traffic load ρ within the capacity region defined in 2), we have S H k ρ d I l), l E. 4) Thi upper bound will be ued to quantify the throughput guarantee of the regulated maximal cheduling policy in the next ection. III. PERFORMANCE OF THE REGULATED MAXIMAL MATCHING SCHEME In thi ection, we how that the network i table under the regulated maximal cheduling when the offered load atifie a pecific condition. We aume that a ρ + kɛ)regulator i employed at kth hop on the route of cla uer. It i worth to mention that the following tability reult i imilar in pirit to that in [9], although there i an important difference here. In fact, we capture flow dynamic in thi paper while the author in [9] only conidered dynamic at the packet level. In [5] and [2], the author captured flow dynamic but their tability reult were achieved by a crolayer congetion control algorithm. In thi paper, network tability i achieved by a imple regulator implementation o communication overhead involved in the congetion control operation can be avoided. The tability reult i tated in the following propoition. Propoition : If the traffic load atifie S H k ρ <, l E 5) then the network i table under the regulated maximal cheduling policy. Thi condition will be called a contraining bound in the equel. Proof: The proof i in Appendix A. Thi reult wa alo tated in [2] where the network wa tabilized by a crolayer congetion control algorithm. Note that thi contraining bound i tight in the ene that the network can become untable with any arbitrarily mall increae of the bound i.e., the right hand ide of 5) become
4 4 + κ for any κ > 0). Thi fact wa proved in everal paper for example, ee [9], [20]). In the following, we tate the performance guarantee for the regulated maximal cheduling policy. Lemma 2: The regulated maximal cheduling policy achieve /d I G) capacity region. Proof: The proof follow directly by comparing the upper and contraining bound on capacity region in 4) and 5), repectively, and uing the definition of d I G). IV. DISTRIBUTED SCHEDULING ALGORITHM A mentioned before, maximal cheduling i a general rule whoe implementation i not pecified. In thi ection, we preent a ditributed cheduling algorithm which approximate the maximal cheduling policy in each time lot within an arbitrarily mall error. In fact, the propoed algorithm will include the BPSIM cheduling algorithm [7] propoed for the node excluive i.e., onehop) interference model a a pecial cae. Our propoed algorithm work with the general k hop interference model. Alo, in contrat to the exiting queuelengthbaed cheduling algorithm [6], [7], in our algorithm each node with incident backlogged link doe not require queue length information of other link in it neighborhood to contruct the tranmiion chedule. In addition, the propoed algorithm i fully ditributed and it ha contant time overhead which doe not grow with the network ize. Our propoed algorithm i, therefore, much more flexible and general than exiting one in the literature. For eae of reference, we will refer to our cheduling algorithm a random approximate maximal matching RAMM) cheduling in the equel. Link are added to the chedule in each round through a matching requet and matching acknowledgment meage exchange a follow. At the beginning of each round, each active node the notion of active/inactive node will be clarified hortly) decide to be left or right with probability /2. Node becoming right wait to receive matching requet from their neighboring node. Backlogged link are added to the chedule in each round a follow. Each left node with at leat one backlogged outgoing link i.e., a link from thi node to one of it neighbor) will chooe a random backoff in [, B]. When the backoff expire, a left node will chooe one of it backlogged neighbor randomly to end a matching requet if it ha not heard any matching requet tranmitted by other node o far in the round. A right node which receive a matching requet will reply with a matching acknowledgment meage C2 C and the correponding link i added to the chedule. We aume that if two or more matching requet are tranmitted in one minilot, colliion occur and no matching acknowledgment C3 A C6 B C5 meage C4 i tranmitted. A. Algorithm Decription The RAMM algorithm i run in the firt phae of each time lot. Specifically, we divide each time lot into two phae: a cheduling phae and a data tranmiion phae. The tranmiion chedule i contructed in the cheduling phae, and it i ued to tranmit data in the data tranmiion phae. The cheduling phae i further divided into K round each of which contain B minilot. In each round, new link are added to the current tranmiion chedule. The tranmiion chedule obtained at the end of the Kth round will be ued to tranmit data in the data tranmiion phae. In addition, only wirele link whoe queue length are larger than the link capacity Scheduling phaetranmiion Time lot are cheduled phae Time lot by the algorithm in each time lot. The time lot tructure of the RAMM algorithm i illutrated in Fig. 2.. Scheduling round Fig. 2. Timing diagram of the RAMM cheduling algorithm. Fig. 3. Illutration of the contention reolution of RAMM cheduling under twohop interference model. In each round, we require that if a link i added to the tranmiion chedule, all wirele link in it interference et be not added to the tranmiion chedule in ubequent round. Thi requirement guarantee that we will obtain a conflictfree tranmiion chedule at the end of the cheduling phae. It i oberved that thi requirement can be eaily achieved by onehop and twohop interference model. Specifically, for the onehop interference model after a link i added to the chedule, both it tranmitting and receiving node will not tranmit and reply any matching requet. For the twohop interference model after a link i added to the chedule, all their onehop neighboring node i.e., one hop away) of both tranmitting and receiving node will be aware of thi through hearing the matching requet or matching acknowledgment) and will not tranmit or reply any matching requet until the end of the cheduling phae. For a general khop interference model with k 3, we aume that a large enough power level i ued to tranmit matching requet/acknowledgment meage in the cheduling phae o that if a link i added to the chedule in one round, all node within k hop from both the tranmitting and receiving node of the link are aware of thi o they will not
5 5 tranmit or reply any matching requet in ubequent round. Node within k hop from the tranmitting and receiving node of any link in the chedule are called inactive node. All other node are active one. Note that any inactive node will remain inactive until the end of the cheduling phae. In general, the number of node participating in the chedule contruction proce reduce rapidly over conecutive round. Becaue new link are kept added to the exiting chedule in each round, the tranmiion chedule in the lat round would approximate well the maximal chedule if B and K are large enough. We will how the performance guarantee of the propoed cheduling algorithm in the next ubection. The contention reolution of the RAMM cheduling algorithm under twohop interference model i illutrated in Fig. 3. In thi figure, if link AB i added to the chedule, all node C i i =, 2,, 6) will be aware of thi through hearing either the matching requet or matching acknowledgment meage from A or B. Hence, after link AB i added to the chedule, thee node C i, i =, 2,, 6) will become inactive. Conequently, all the link in thi figure except AB which are conflict with link AB will never be added to the chedule in the ubequent round. B. Analyi Now, let degree d i of node i be the number of node having link directly connecting to node i i.e., onehop neighbor of node i). Let d be the maximum of d i for all node in the network i.e., d = max i V d i ). In addition, a matching requet tranmitted by one node may collide with thoe tranmitted by other node. Let I i be the number of node whoe tranmitted matching requet may collide with that of node i if node i and one or more of thee node tranmit imultaneouly under the correponding power level ued in the cheduling phae. Let I be the maximum of I i i.e., I = max i V I i ). Alo, let I0 be the maximum number of node which are at mot k hop away from either A or B including A and B for any link AB in the network. We have the following reult. Propoition 2: For any µ 0, ), we can chooe the number of cheduling round K which depend only on B, d, I, I 0, and µ but independent of network ize uch that for any backlogged link l, the probability that at leat one backlogged link in it interference et I l i cheduled after K round i larger than or equal to µ. Proof: The proof i in Appendix B. Uing RAMM cheduling algorithm together with regulator implementation a decribed in ection II, we have the following tability reult. Propoition 3: If the traffic load atifie S H k ρ < µ, l E 6) and under the condition tated in propoition 2, the network will be table when RAMM algorithm i ued together with the regulator implementation a decribed in ection II. Proof: The proof follow the ame line with that of Propoition. However, the right hand ide of the contraining bound become µ intead of one due to the performance bound achieved by RAMM cheduling cheme. The reult in thi propoition mean that we can achieve the performance bound of the regulated maximal matching tated in propoition within an arbitrarily mall error by uing the RAMM cheduling algorithm with contanttime overhead. V. LONGTERM UTILITY MAXIMIZATION UNDER LOW LOAD CONDITION Propoition 3 alo mean that when regulator and RAMM cheduling algorithm are implemented and traffic load atifie the condition tated in 6), the network i table a long a uer rate are bounded away from zero. A a conequence of thi reult, it i clear that we do not need any congetion control algorithm a long a the traffic load in the network i low. Hence, communication overhead due to meage exchange of the congetion control algorithm can be avoided if regulator are implemented in the network. In addition, the number of uer for each cla change dynamically due to the flow level dynamic, o intantaneou network utility maximization may not lead to good network performance. Therefore, under thi tability condition, it i natural to ak: how to chooe uer rate uch that optimal longterm time average) network utility can be achieved? Specifically, our objective i to maximize the longterm network utility which can be explicitly tated a max xt) lim τ τ τ t=0 [ S ] n t t)u x t)) dt 7) where n t t) and x t) are the number of cla uer tranmitting in time lot t and their tranmiion rate, repectively; U x ) i the utility function, which can, for example, reflect the level of atifaction for cla uer. We aume that uer arriving during time lot t can only tranmit from time lot t + onward. Suppoe that the queueing proce at each ource node i ergodic thi fact wa jutified in [25]). Let f n t, x) denote the joint probability denity function of n t and x in equilibrium. Becaue element [ of n t are pairwie S ] independent, we have f n t, x) = fnt x) f x). Thu, we can rewrite 7) a S max n t U x )fn t x) f x)d x. 8) x X n t =0 Let u define g x) = = = = S n t U x )fn t x) n t =0 S U x ) n t fn t x) n t =0 S U x )E [ N x t ] S ρ x U x )
6 6 where we have ued Little law in deriving E [N x] t in the above equation. Specifically, the expected waiting time for a cla uer i /µ x ), uing Little law we have E [N x] t = λ /µ x ) = ρ /x. Thu, we can rewrite 8) a max x g x)f x)d x 9) X Now, uppoe we wih to find optimal uer rate x [0, M ]. Let x be the maximum of g x ) = ρ /x U x ) in [0, M ] and the correponding optimum rate vector i x. Then, it i eay to ee that chooing f x) = δ x x ) will maximize 9) where δ.) i the delta function. Thu, the longterm utility maximization can be achieved by allowing uer of cla to tranmit at the optimal rate x. In ummary, when the traffic load atifie 6), the network i table under the propoed cheduling policy and no congetion control i needed. In addition, we can decouple the longterm utility maximization from tability under thi tability condition. Example: When the utility function i U x ) = lnx ) which correpond to proportional fair rate allocation among uer, we have g x ) = ρ /x lnx ). The global maximum of g x ) i x = e. Thu, if M > e, the optimal tranmiion rate to achieve maximum longterm utility i x = e. We will compare longterm utility under thi olution and for the cae where crolayer congetion control algorithm i ued [2]. VI. CONGESTION CONTROL UNDER HEAVY LOAD In thi ection, we conider the heavy load regime where the bound tated in 5) i violated. Thi may be the cae when there are many longlived flow in the network. A. CroLayer Congetion Control Algorithm In thi ubection, we preent a crolayer congetion control algorithm which can tabilize the network under any offered load. We aume that the flow dynamic are low compared to the time cale of a congetion control algorithm. Hence, we aume that there are fixed number of uer of each cla in the network which will be denoted by N. Our propoed crolayer congetion control algorithm work a follow. where U.) i the invere of derivative of utility function U.) and [x] + = max[x, 0]. The implicit cot are updated by q l t + ) = [ q l t) + α l S H k N x + κ )] + where κ > 0 i a mall number and α l i the tepize. Tranmiion cheduling: The regulated maximal cheduling policy i employed in each time lot. We aume that the utility function U x ) i increaing, trictly concave, twice differentiable. The propoed congetion control algorithm implicitly olve the following optimization problem. S maximize N U x ) ubject to S H k N x κ, l E 0) Note that contraint of thi optimization problem come from the contraining bound tated in 5). Here, the tranmiion rate of cla uer i ued intead of it offered load. We introduce a mall κ > 0 on the right hand ide of thee contraint to enure that feaible olution of thi optimization problem trictly atify the contraining bound. By introducing Lagrange multiplier q l for the contraint in 0), we have the following Lagrangian L x, q) = S N U x ) l E q l [ S H k N x + κ Uing the tandard dual decompoition technique, we can calculate the uer rate from x t) = argmax N U x ) N x q l t) 0 x M l E :H k= which give the uer rate update rule preented in the algorithm. Alo, implicit cot update tep of the propoed algorithm come from the dual update of the underlying optimization problem. Now, let x be the optimal olution of the preented algorithm, we implement a x + kɛ)regulator in the kth hop of cla uer. We have the following reult on the tability of the preented algorithm. Propoition 4: If the tepize i choen to be mall enough, the preented algorithm can tabilize the network under any offered load. Proof: The proof i in Appendix C. It i expected that a congetion control algorithm hould tabilize the network under any offered load. In thi repect, the propoed congetion control algorithm doe a good job compared to thoe in [5], [2]. Note that in practice where the RAMM cheduling algorithm i implemented to approximate Uer rate i determined by the maximal matching policy RAMM intead of maximal x t) = min U + matching i ued in tep 3 of thi algorithm), we can modify q l t), M the propoed congetion control algorithm by replacing on the right hand ide of the contraint in 0) by µ and modify l E :H k= the algorithm accordingly. B. Some Implementation Iue In practice, there i no imple way to know whether or not the contraining bound in 5) i atified. Therefore, it i unknown when the congetion control algorithm propoed in ection VI.A hould be activated. Moreover, it would be wie to avoid performing congetion control if it i unneceary. Alo, uer in the network hould tranmit at their optimal rate a calculated in ection V to maximize the longterm ]
7 utility if the network i known to be table by the regulated maximal cheduling policy. To achieve uch goal, uer would attempt to tranmit at their optimal rate auming that the offered load i low and the network i table. If thi i not the cae, the network hould be equipped with an appropriate mechanim to detect and inform all uer about the ongoing congetion and requet them to activate the congetion control algorithm. Congetion can be detected at network node through oberving frequent buffer overflow and/or large endtoend delay. Upon detecting network congetion, network node hould end appropriate feedback to inform all uer in the network and requet them to run the congetion control algorithm. Thee operation enure that the bet performance can be achieved and the mot appropriate operation are performed at all time. Fig Grid network of 36 node with 30 onehop flow or 6 multihop flow. 7 VII. NUMERICAL RESULTS In thi ection, we how ome illutrative numerical reult for the propoed cheduling algorithm and the longterm utility maximization. We conider grid network with onehop and multihop flow. We aume that tranmiion rate on each wirele link equal = 0 unit/time lot, average length of each file brought by any uer cla i /µ = 0 unit. Uer of each cla arrive according to Poion proce with arrival rate λ. We vary arrival rate to adjut the traffic load ρ = λ /µ Here, a unit of data i a block of information bit of uitable ize We aume all 22 7 flow 23 8 have 24 the ame load ρ in all the reult Fig. 4. Grid network of 36 node with 60 onehop flow or 2 multihop flow. A. Performance of Scheduling Algorithm In Fig. 7, we how the minimum probability of achieving a maximal chedule due to the RAMM algorithm veru the number of round K under different maximum backoff value B for the grid and random topologie hown in Fig. 4, Fig. 6 under the onehop interference model. For both topologie, we aume there are two flow in two different direction on each link i.e., for any link AB, there are one flow from A to B and one flow from B to A). We aume that all the flow are alway backlogged which i the wort cae cenario. The probability of achieving a maximal chedule for a backlogged Fig. 6. Random network of 36 node with 98 bidirectional link 99 undirectional link), d = 9, I = 24. link i the probability that it i cheduled or at leat one link in it interference et i cheduled. The minimum matching probability i the mallet probability for all link in the network. We have obtained thi probability by averaging over 0 4 time lot. It i oberved that with B = 3, to achieve a minimum matching probability of 0.9, we need K = 4 and 5 round for the grid and random topologie, repectively. Alo, with B = 3 and K = 9, i.e., the total number of minilot required i M = B K = 27), the minimum matching probability i almot. Similarly, we how the minimum probability of achieving a maximal chedule due to the RAMM algorithm veru the number of round K under different maximum backoff value B for the grid and random topologie under twohop interference model in Fig. 8. Thi figure how that with B = 4, we only need K = for the grid topology and K = 3 for the random topology to achieve minimum matching probability very cloe to. It i alo hown that even the ize of the node interference et i very large I = 22 for the grid topology and I = 24 for the grid topology), the improvement of minimum matching probability i very marginal when the maximum backoff B i larger than 8. In fact, it i more beneficial in term of overhead i.e., M = B K) to ue mall value of B e.g., B = 4 or 5 i a good choice).
8 8 Minimum matching probability Grid, B = 2 Random, B = 2 Grid, B = 3 Random, B = 3 Grid, B = 4 Random, B = Number of round K Total average queue length 9 x RAMM, M = 20 RAMM, M = 28 RAMM, M = 36 RAMM, M = 44 GMM P&C, M = Load ρ Fig. 7. Minimum probability of achieving the maximal cheduling veru number of round for a grid topology in Fig. 4 with 20 onehop flow or random topology in Fig. 6 with 98 onehop flow under onehop interference model. Fig. 9. Performance of RAMM cheduling algorithm under twohop interference model for maximum backoff value B = 4, grid network with 36 node and 30 onehop flow in Fig. 5). Minimum matching probability Grid, B = 4 Random, B = 4 Grid, B = 5 Random, B = 5 Grid, B = 8 Random, B = 8 Grid, B = 0 Random, B = Number of round K Total average queue length 9 x Policy W, M = 44 Policy W, M = 28 Policy W, M = 52 Policy W, M = 024 GMM P&C, M = Load ρ Fig. 8. Minimum probability of achieving the maximal cheduling veru number of round for a grid topology in Fig. 4 with 20 onehop flow or random topology in Fig. 6 with 98 onehop flow under twohop interference model. B. Comparion of Different Scheduling Algorithm In thi ubection, we compare performance of the propoed RAMM cheduling algorithm with the other two queuelengthbaed contanttime cheduling algorithm in [6], namely policy V and policy W for onehop and twohop interference model, repectively. For eae of reference, we briefly decribe thee two cheduling algorithm here. Before doing o, let u introduce ome neceary notation. Let l) and rl) denote the tranmitting node and receiving node of link l, repectively. Alo, let Ei) denote the et of link connected to node i, and Nl) denote the et of neighboring link haring a common node with link l i.e., Nl) = El)) Erl))\{l}. In addition, let Nl) + denote the union of Nl) and {l}, and n l and n + l denote the number of link in Nl) and Nl) +, repectively. For the comparion purpoe, we aume inglehop flow o data traffic will be directly buffered in the tranmiion queue for each link i.e., no regulator implementation). In the cheduling policie V and W, each time lot alo conit of cheduling phae and data tranmiion phae. There are Fig. 0. Performance of policy W [6] under twohop interference model for M = 44, 28, 52, 024, grid network with 36 node and 30 onehop flow in Fig. 5). M minilot in the cheduling phae. The two cheduling algorithm work a follow. Policy V: In each minilot of the cheduling phae, each backlogged link contend with probability α x l M where α = M )/2 and Q l / x l = { max k El)) Q k/, } k Erl)) Q k/ where Q l i the queue length and i the tranmiion rate of link l, repectively. Policy W: In each minilot of the cheduling phae, each backlogged link contend with probability β y l M where M β = n ) where n = max l E n + l y l = max k Nl) + and Q l /. h Nk) Q + h /R h In Fig. 9, we how the performance of RAMM cheduling algorithm in a grid network with 36 node and 30 onehop
9 9 Total average queue length x 05 RAMM, M = 20 RAMM, M = 28 RAMM, M = 36 RAMM, M = 44 GMM Load ρ Fig.. Performance of RAMM cheduling algorithm under twohop interference model for B = 4, grid network with 36 node and 6 multihop flow in Fig. 5). Total average queue length x 05 Policy V, M = 28 Policy V, M = 52 Policy V, M = 024 RAMM, M = 27 GMM Load ρ Fig. 2. Performance of RAMM with B = 3, K = 9) and policy V [6] cheduling algorithm under onehop interference model for grid network with 36 node and 60 onehop flow in Fig. 4). flow a in Fig. 5. For the RAMM algorithm, we how the performance with different tranmiion round K while we fix the maximum backoff value of B = 4 the number of minilot in the cheduling phae i M = B K). We alo preent performance of PickandCompare P&C) [], [3] and greedy maximal matching GMM) [5] cheduling cheme. For the P&C cheme, a new chedule i generated in each time lot by uing RAMM algorithm. Then the total weight of the old chedule and the newly generated chedule are compared where the weight of a link i the length of it tranmiion queue, and the chedule with larger weight i choen for tranmiion in the time lot. In fact, the huge complexity of the P&C cheme incur in the compare tep. For the GMM cheme, the chedule i contructed by adding one feaible link with the larget weight to the chedule and removing all conflicting link with the added link in each tep. Thi procedure i repeated until no further link can be added to the chedule. Note that both P&C and GMM cheduling cheme are either centralized or require huge overhead to implement in a ditributed manner. Alo, it i known that thee two cheduling cheme achieve almot the capacity region. It i confirmed in thi figure that P&C and GMM cheme achieve imilar performance although GMM ha a bit maller total queue length for traffic load cloe to the boundary of the capacity region. It i evident that when we increae the number of round for the RAMM algorithm, we achieve better performance. Moreover, in the conidered network d I G) = 4, o the performance guarantee tated in propoition i jut /4 capacity region. However, the actual performance achieved by the propoed RAMM cheduling algorithm i much better than the performance bound a can be een in Fig. 9. In Fig. 0, we preent the performance of the queuelengthbaed contant time cheduling policy W for the two hop interference model [6]. We how the performance of thi cheduling cheme for different number of minilot ued in the cheduling phae M the firt phae of thi cheduling policy correpond to the firt phae of our RAMM cheduling algorithm). A i evident, with M = 44 minilot, policy W achieve much lower performance than the RAMM algorithm. Even with M = 024 minilot, performance of policy W i till wore than that of RAMM algorithm with only M = 44 minilot. Note that due to practical implementation contraint, the time lot duration i uually limited to few milliecond a in mot current wirele ytem. If the duration of a minilot i 20 µ a in the WLAN tandard, M = 00 correpond to 2 m which i already quite large. Our cheduling algorithm, therefore, preent a ignificant advantage compared to policy W becaue we cannot make the time lot interval arbitrarily large in practice. In Fig., we how performance of the RAMM cheduling algorithm for the grid network with 6 multihop flow in Fig. 5. Similar performance to the inglehop cae preented in Fig. 9 i oberved for thi etting. It i evident that although multihop flow contribute more traffic to the network becaue each flow travere multiple link, cheduling algorithm i till the key to determine performance of the network. In Fig. 2, we compare performance of RAMM and policy V [6] under onehop interference model for the grid network with 60 onehop flow in Fig. 4. It i evident that RAMM cheduling with only M = 27 minilot achieve imilar performance with policy V for M = 024 minilot. The RAMM cheduling algorithm, therefore, ha ignificantly lower overhead compared to policy V. Note that the RAMM cheduling algorithm doe not require queue length information, o it i much eaier to implement compared to policy V and W. In addition, collecting queue length information will create further overhead which may alo be very ignificant, epecially for the twohop interference model becaue each node need to forward queue length information of it incident link two hop away). C. Longterm Utility We compare the longterm time average) utility under the optimal tranmiion rate derived in ection V and under the cae where a crolayer congetion control algorithm i ued. Specifically, we will conider the crolayer congetion control algorithm propoed in [2] which i the extenion of
10 0 that for onehop interference model in [5]. The crolayer congetion control algorithm work a follow: Congetion price for each link l i updated a q l t + ) = [q l t) + α l q l t)] + 2) where α l i the tep ize and q l t) = [ S k Il) H k t+ t n t)x t) k St) and St) denote the et of link belonging to the chedule in time lot t,.) i the indicator function. Tranmiion rate of cla uer i updated a l E q lt + ) H k k Il) x t + ) = min, M. 3) Tranmiion cheduling: The network i cheduled in each time lot by the correponding cheduling algorithm GMM or RAMM algorithm). Average utility Congetion control, RAMM Congetion control, GMM Optimal tranmiion rate Load ρ Fig. 3. Average utility of the propoed optimal tranmiion rate and with congetion control algorithm under onehop interference model for grid network with 36 node and 2 multihop flow, GMM and RAMM cheduling with B = 3, K = 9 in Fig. 4). We conider the utility function U x ) = lnx ). Hence, for the propoed approach the optimal tranmiion rate for each uer i x = e i.e., we aume that M > e). We will illutrate performance of the crolayer congetion control algorithm under both GMM and RAMM cheduling algorithm. Longterm average utility i obtained by averaging the utility over time lot. For the crolayer congetion control algorithm, we fixed the tranmiion rate at x = e for the firt 0 3 time lot while till updating the congetion price and generating a new chedule in each time lot. Thi initial period provide time for the congetion price to converge to better value. The tep ize i initialized a α = 0. and it i updated a α = max { α/t, 0 3}. We how the average utility achieved by the optimal tranmiion rate and by the congetion control algorithm in Fig. 3, 4 for onehop and twohop interference model, repectively. It i evident that the propoed approach achieve higher average utility then thoe uing congetion control ] Average utility Congetion control, RAMM Congetion control, GMM Optimal tranmiion rate Load ρ Fig. 4. Average utility of the propoed optimal tranmiion rate and with congetion control algorithm under twohop interference model for grid network with 36 node and 6 multihop flow, GMM and RAMM cheduling with B = 4, K = in Fig. 5). algorithm for all traffic load under both interference model. In fact, the average utility with congetion control decreae ignificantly when the traffic load i cloe to the boundary of the region which can be tabilized by the underlying cheduling algorithm. Thi obervation confirm the argument that performing congetion control i unneceary if the network can be tabilized by the underlying regulated cheduling algorithm. VIII. CONCLUSIONS We invetigated the network control problem uing contanttime cheduling under the khop interference model. With flow dynamic conideration, we have hown that the network can be tabilized by uing a regulated maximal cheduling policy if the offered load atifie the contraining bound. We have preented a contanttime and ditributed cheduling algorithm for a general khop interference model. The cheduling algorithm doe not require queue length information and ha overhead not growing with network ize. Our propoed cheduling algorithm achieve performance arbitrarily cloe to that of the regulated maximal cheduling. Under the tability condition, we have derived optimal tranmiion rate which achieve the maximum longterm network utility. For the high load regime, we have propoed a crolayer congetion control algorithm which can tabilize the network for any offered load. Numerical reult have hown that the propoed cheduling algorithm achieve much better performance than the exiting contanttime cheduling algorithm in the literature, and it ha much better performance than it performance guarantee. Alo, performing congetion control under low load condition actually degrade performance in term of longterm utility ignificantly compared to the optimal tranmiion rate. APPENDIX I PROOF OF PROPOSITION Let Q l t) and Q lt) be tranmiion queue length for cla and for all uer clae at link l in time lot t, repectively. Similarly, let P l t) and P lt) be regulator queue length
11 for cla and for all uer clae at link l in time lot t, repectively. Let u denote by b l and a l the previou link and next link of link l on the route of cla uer. Alo, let Cl t) and Dl t) be the number of packet tranmitted from regulator and tranmiion queue in time lot t, repectively. We have the following queue update equation Q l t + ) = Q l t) Dl t) + Cl t) 4) Pl t + ) = Pl t) Cl t) + Db t). l 5) Thu, we have Q l t+)+pa t+) = Q l l t)+pa t)+c l l t) Ca t). 6) l We will ue the following Lyapunov function for the ytem where V P, Q) = V Q) + ξv 2 P, Q) 7) V Q) = l V 2 P, Q) = l Q l t) Q k t) P a l + Q l ) 2. In fact, thi Lyapunov function wa alo ued in [9]. Now, let u conider V Qt + )) V Qt)) = 2 ) Q l t) Q k t + ) Q k t)) l + ) Ql t + ) Q l t)) l ) Q k t + ) Q k t)) = 2 l + l Q l t) ) Ck t) D k t)) C l t) D l t)) C k t) D k t)) ) 8).9) Since the number of packet tranmitted from regulator and tranmiion queue in each time lot are bounded, the econd term in 9) can be bounded by a contant B. Thu, we have [ E V Qt + )) V Qt)) P t), Qt) ] 2E [ l Q l t) 2E Q l t) l:q l t) C k t) D k t)) C k t) ) P t), Qt) ] ) D k t) + B P t), Qt) ] + B 2. Let L be the larget number of hop travered by any uer cla, we have E Q l t) Ck t) P R t), Qt) l:q l t) k Q l t) H k ρ + Lɛ. 20) R l:q l t) k Alo, due to the definition of maximal cheduling policy, we have Q l t) D k t) Q l t). 2) l:q l t) l:q l t) For traffic load atifying 5), we can find ɛ and δ mall enough uch that S H k ρ + Lɛ + δ, l E. 22) From 20), 2), 22), we have [ E V Qt + )) V Qt)) P t), Qt) ] Q l t) δ + B 2 δ l:q l t) l Q l t) + B 3. 23) Now, uing the procedure a in [9], we can obtain [ E V 2 P t + ), Qt + )) V 2 P t), Qt)) P t), Qt) ] Q l t) η Pa t) + B 4. 24) l l l Combining the reult in 23), 24), we have [ E V P t + ), Qt + )) V P t), Qt)) P t), Qt) ] [ ] δ ξ Q l t) ξη Pa R t) + B 5. 25) l l l l δ We can chooe ξ mall enough uch that ξ > 0. Thu, the drift will be negative if the regulator and/or tranmiion queue become large enough. Therefore, the tability reult follow by uing theorem 2 of [26]. Note that the choen Lyapunov function doe not take regulator queue on the firt hop of each uer cla into account. Thee regulator queue are, however, table becaue their output rate i ρ + ɛ which i larger than the average input load i.e., ρ ). APPENDIX II PROOF OF PROPOSITION 2 Conider any link AB between node A and B. We will find the probability that at leat one link in the interference et I AB i cheduled. Thi event will be denoted a M AB in the equel. A mentioned before, RAMM cheduling algorithm include BPSIM cheduling algorithm for the onehop node excluive) interference model propoed in [7] a a pecial cae. In the following we prove propoition 2 for k 2. For the pecial cae of k =, we refer the reader to [7] for the proof and the correponding analyi. Note that the proof for the cae k 2 i very challenging and completely different from that for the
12 2 cae k = in [7] due to the more complicated interference relationhip. We will illutrate ome important definition ued in the proof in Fig. 5. Let I 0 be the et of node which i at mot k hop away from either A and B including A, B. For the grid network and link AB hown in Fig. 5 under the twohop interference model, I 0 = {A, B, C,, C 6 }. Alo, the interference et for node C 6 i.e., I C6 ) conit of all node in I 0 and all blank node. Note that the notion of node interference et i different from that of link interference et provided in definition I. Here, all node in I C6 are at mot 3 hop away from C 6. We oberve that all link incident to any node in I 0 will belong to I AB becaue they are within k hop from link AB. In the following, we will find the lower bound for the probability of M AB by conidering ubcae in which there are i left node in the et I 0. For convenience, we will ue I 0 to denote both the et itelf and the correponding number of node in I 0. Let L i be the event that there are i left node in the et I 0. The probability that at leat one link in the interference et of link AB i cheduled can be lower bounded a P m I 0 i= PrM AB L i )PrL i ). 26) Recall that for any particular node A, there are at mot I node whoe matching requet can collide with that tranmitted from node A with the power ued in the cheduling phae. Note that thee I node will be at mot k + hop away from mode A for the interference model with k 2 where the ame tranmiion power level i ued in both cheduling and tranmiion phae of each time lot. To find the lower bound for the probability of M AB, we will aume the wortcae cenario where each node ha I interfering node. For convenience, we alo ue I A to denote the et of thee interfering node whoe tranmiion can collide with that from node A. We now C conider C6 the following cae. C2 C3 A C4 B C5 Fig. 5. Interference et I C6 under twohop interference model. A. There i only one left node in I 0 We conider the following two ubcae. If thi left node i either A or B, then the left node will have at leat one neighbor which i a right node. Thi cae occur with probability 2/2) I0. For eae of reference, we will refer to thi left node a node C i.e., C i either A or B). Alo, there are at mot I I 0 node whoe matching requet can collide with that from node C. To find the lower bound for PrM AB ), we aume that there are I I 0 uch interfering node. Now, uppoe there are i left node among thee I I 0 interfering node. The matching requet tranmitted by node C will be uccefully received if the backoff value of thee i left node are larger than that of node C. Specifically, the matching requet from node C will be uccefully received and the correponding link will be cheduled with a probability which i lower bounded by F = I I 0 i=0 I I0 i ) ) I I0 2 B B m= m ) i B where we have broken the event into ubcae where there are i left node among I I 0 interfering node and thee i left node have backoff value larger than that of node C. If thi left node i any node other than A and B then it can be any node among I 0 2) node. Again, we refer to thi left node a node C. Note that all node which are within one hop from A or B including A and B are at mot k + hop away from C o they all belong to the interfering et I C. Let x be the total number of node within one hop from A and B including A and B, then there are at mot I x node whoe matching requet can collide with that from node C. Thi i becaue thee x node belong to I 0 and they are all right node except C if C i in I 0. We will aume thi wortcae cenario to calculate the lower bound of the matching probability in the following. Note that node C will have at leat one neighbor which i a right node becaue it i the only left node in I 0. Again, uppoe there are i left node among potential interfering node in I C. The matching requet tranmitted by node C will be uccefully received if the backoff value of thee i left node are larger than that from node C and the matching requet from node C i tranmitted to a right node. Now, we conider the following two ubcae. For the firt cae, if C i one of x node i.e., x onehop neighbor of A or B) but not A and B. Then, thi cae occur with probability x 2)/2) I0. In thi cae, the matching requet from node C will be uccefully received and the correponding link will be cheduled with a probability which i lower bounded by F 2 = I x i=0 I x i ) 2 ) I x B B m= m ) i B where i i the number of left node. Alo, in calculating the lower bound for P m we aume that C tranmit it
13 3 matching requet to a right node which i not an onehop neighbor of A or B. Hence, there are at mot I x left node which can collide with the matching requet from C. For the econd cae, if C i not one of x node i.e., x onehop neighbor of A or B). Then, thi cae occur with probability I 0 x)/2) I 0. In thi cae, the matching requet from node C will be uccefully received and the correponding link will be cheduled with a probability which i lower bounded by F 3 = I x 2 i=0 I x 2 i ) ) I x 2 B B m= m ) i B where in calculating the lower bound for P m we aume that C tranmit it matching requet to a right node which i not an onehop neighbor of A or B. Hence, there are at mot I x 2 left node which can collide with the matching requet from C. B. There are two or more left node in I 0 Suppoe node C in the et I 0 become left and win the contention. Then node C hould have the mallet backoff value among all the node whoe matching requet can collide with the matching requet from C. Alo, node C hould end the matching requet to a node which i a right one. For eae of reference, we will refer to thi right node a node D in the equel. In general, D can belong to et I 0 or not. However, to find the lower bound of P m, we aume that D belong to I 0 ; therefore, there are at mot I 0 2 other left node beide C and D in I 0. A before, we aume the wortcae cenario where there are I node whoe tranmiion can collide with that of node C. Recall that all x node which are onehop neighbor of A or B belong to the et I. Similar to the previou cae, we conider the following two ubcae. For the firt cae, C i one of x node i.e., x onehop neighbor of A or B). In thi cae the matching probability can be lower bounded a F 4 = x I 0 2 i= I x j=0 I0 2 i ) ) I0 I x 2 j ) I x 2 B B m= ) m ) i+j B where i i the number of left node beide C and D in the et I 0. And j i the number of left node which belong to I but are not onehop neighbor of A or B i.e., there i no link between thee node and A or B). We will denote thi et a I \ x in the equel. In general, D can belong to I \ x or not; however, to find the lower bound for P m, we only allow j take value from 0 to I x. In addition, C can be any node among x node o we have a factor of x before the um. Alo, we require that all left node i left node belonging I 0 and j left node belonging to I \ x) achieve larger backoff value than that of node C. For the econd cae, C belong to the et I \ x. In thi cae, the matching probability can be lower bounded a F 5 = I 0 x) I 0 2 i= I x 2 j=0 I0 2 i ) ) I0 I x 2 2 j ) I x 2 B B m= ) m ) i+j B where C can be one of I 0 x node o we have the factor I 0 x) before the um. Alo, to calculate the lower bound for the matching probability, we aume D alway belong to I \ x, o j can be at mot I x 2. Subtitute reult of all conidered cae into 26), the matching probability i lower bounded by P m P 0 m = 2/2) I 0 F + x 2)/2) I 0 F 2 +I 0 x)/2) I 0 F 3 + F 4 + F 5 27) where F, F 2, F 3, F 4, and F 5 are defined above. From the lower bound of the matching probability P 0 m derived above, we can calculate the lower bound of P m for any link AB a p = min x, I 0 P 0 m where we find the minimum of P 0 m over all poible x and I 0. Note that poible value of x and I 0 will be in the range of [3, 2d ] and [3, I 0 ], repectively. Here, x and I 0 are at leat three for the network to be connected i.e., A and B hould have at leat one onehop neighbor). It i oberved that p i independent of the network ize and depend only on B, d, I, I 0. Now, we can chooe K uch that at leat one link in I l of a backlogged link l i cheduled with probability greater than µ after K round a follow: { K = min k : p ) k µ }. 28) k Thu, we can chooe K which i independent of network ize and only depend on B, d, I, I 0, µ uch that performance guarantee arbitrarily cloe to the contraining bound can be achieved. Example: For the grid network and twohop interference model, we have I = 22, I 0 = x can take value of 5, 6, 8. With maximum backoff value B = 0, by uing the analyi preented above, the minimum number of cheduling round to achieve µ = 0.9 i K = 60. In fact, thi calculation i quite conervative becaue it conider the wort cae cenario. In practice, the ize of interference et decreae quickly over cheduling round, o the required value of K i much maller. APPENDIX III PROOF OF PROPOSITION 4 The proof i imilar to that of Propoition i.e., we ue the ame Lyapunov function and proof procedure). In particular,
14 4 we have a imilar bound a in 20) a follow: E Q l t) Ck t) P R t), Qt) l:q l t) k Q l t) H k x + Lɛ. 29) R l:q l t) k Due to the contraint of the optimization problem in 0), we can find ɛ and δ mall enough uch that S H k x + Lɛ + δ, l E. 30) From 29) and 30), we have [ E V Qt + )) V Qt)) P t), Qt) ] Q l t) δ + B 6 δ l:q l t) l Q l t) + B 7. 3) The remaining tep to obtain negative drift when regulator and/or tranmiion queue become large enough are the ame a in the proof of Propoition. REFERENCES [7] A. Gupta, X. Lin and R. Srikant, Lowcomplexity ditributed cheduling algorithm for wirele network, IEEE INFOCOM [8] X. Wu and R. Srikant, Bound on the capacity region of multihop wirele network under ditributed greedy cheduling, IEEE INFOCOM [9] X. Wu, R. Srikant, and J. R. Perkin, Scheduling efficiency of ditributed greedy cheduling algorithm in wirele network, IEEE Tranaction on Mobile Computing, vol. 6, no. 6, pp , June [20] P. Chaporkar, K. Kar, S. Sarkar, Throughput guarantee through maximal cheduling in wirele network, Allerton Conference on Communication, Control and Computing, Sept [2] G. Sharma, R. Mazumdar, and N. Shroff, Joint congetion control and ditributed cheduling for throughput guarantee in wirele network, IEEE INFOCOM [22] C. Hume, Jr., A regulator tabilization technique: KumarSeidman reviited, IEEE Tranaction on Automatic Control, vol. 39, no., pp. 996, Jan [23] P. R. Kumar and T. I. Seidman, Dynamic intabilitie and tabilization method in ditributed realtime cheduling of manufacturing ytem, IEEE Tranaction on Automatic Control, vol. 35, no. 3, pp , Mar [24] J. Zhang, D. Zheng, and M. Chiang, The impact of tochatic noiy feedback on ditributed network utility maximization, IEEE INFOCOM [25] K. Ma, R. R. Mazumdar, and J. Luo, On the performance of primal/dual cheme for congetion control in network with dynamic flow, IEEE INFOCOM [26] M. Neely, E. Modiano, and C. Rohr, Power allocation and routing in multibeam atellite with timevarying channel, IEEE/ACM Tranaction on Networking, vol., no., pp , Feb [] J. Mo and J. Walrand, Fair endtoend windowbaed congetion control, IEEE/ACM Tran. Networking, vol. 8, no. 5, pp , Oct [2] F. P. Kelly, A. Maulloo, and D. Tan, Rate control in communication network: Shadow price, proportional fairne and tability, Journal of the Operation Reearch Society, vol. 49, pp , 998. [3] S. H. Low and D. E. Lapley, Optimal flow control, I: Baic algorithm and convergence, IEEE/ACM Tranaction on Networking, pp , Dec [4] H. Yaiche, R. Mazumdar, and C. Roenberg, A game theoretic framework for bandwidth allocation and pricing in broadband network, IEEE/ACM Tranaction on Networking, vol. 8, no.5, pp , [5] X. Lin and N. B. Shroff, The impact of imperfect cheduling on crolayer congetion control in wirele network, IEEE/ACM Tranaction on Networking, vol. 4, no. 2, pp , April [6] M. Neely, E. Modiano, and C. Li, Fairne and optimal tochatic control for heterogeneou network, IEEE INFOCOM [7] B. Hajek and G. Saaki, Link cheduling in polynomial time, IEEE Tranaction on Information Theory, vol. 34, pp , 988. [8] L. Bui, A. Eryilmaz, R. Srikant, and X. Wu, Joint aynchronou congetion control and ditributed cheduling for multihop wirele network, IEEE INFOCOM [9] G. Sharma, R. Mazumdar, and N. Shroff, On the complexity of cheduling in multihop wirele ytem, IEEE MOBICOM [0] L. Taiula and A. Ephremide, Stability propertie of contrained queueing ytem and cheduling policie for maximum throughput in multihop radio network, IEEE Tranaction on Automatic Control, vol. 37, no. 2, pp , Dec [] L. Taiula, Linear complexity algorithm for maximum throughput in radio network and input queued witche, IEEE INFOCOM 998. [2] E. Modiano, D. Shah, and G. Zuman, Maximizing throughput in wirele network via goiping, ACM SIGMETRICS [3] A. Eryilmaz, A. Ozdaglar, and E. Modiano, Polynomial complexity algorithm for full utilization of multihop wirele network, IEEE INFOCOM [4] S. Sanghavi, L. Bui and R. Srikant, Ditributed link cheduling with contant overhead, ACM SIGMETRICS [5] X. Lin and S. Raool, Contanttime ditributed cheduling policie for ad hoc wirele network, IEEE Conference on Deciion and Control, Dec [6] C. Joo and N. B. Shroff, Performance of random acce cheduling cheme in multihop wirele network, IEEE INFOCOM PLACE PHOTO HERE Long Le received the B.Eng. degree with highet ditinction from Ho Chi Minh City Univerity of Technology in 999, the M.Eng. degree from Aian Intitute of Technology AIT) in 2002, and the Ph.D. degree from Univerity of Manitoba in He i currently a Potdoctoral Fellow in the Department of Electrical and Computer Engineering at Univerity of Waterloo. Hi current reearch interet include cognitive radio, cooperative diverity and relay network, and crolayer deign for communication network. He i a Member of the IEEE. Ravi Mazumdar wa born in Bangalore, India. He obtained the B.Tech. in Electrical Engineering from the Indian Intitute of Technology, Bombay, India in 977, the M.Sc. DIC in Control Sytem from Imperial College, London, U.K. in 978 and the Ph.D. in Sytem Science from the Univerity of California, Lo Angele, USA in 983. He i currently a Univerity Reearch Chair Profeor of Electrical and Computer Engineering at the Univerity of Waterloo, Waterloo, Canada and an Adjunct Profeor of ECE at Purdue Univerity. He ha erved on the facultie of Columbia Univerity NY, USA), INRS Telecommunication Montreal, Canada), Univerity of Eex Colcheter, UK), and mot recently at Purdue Univerity Wet Lafayette, USA). He ha held viiting poition and abbatical leave at UCLA, the Univerity of Twente Netherland), the Indian Intitute of Science Bangalore); and the Ecole Nationale Superieure detelecommunication Pari). He i a Fellow of the IEEE and the Royal Statitical Society. He i a member of the working group WG6.3 and 7. of the IFIP and a member of SIAM and the IMS. He hared the Bet Paper Award with G. Sharma and N. Shroff at the INFOCOM 2006 and wa coauthor with N. Likhanov for the Bet Paper runner up at INFOCOM 998. He i an Editor of the IEEE Tran. on Networking and ha erved a a guet editor for a number of journal. Hi reearch interet are in wirele and wireline network; application of game theory to networking; applied probability, queueing theory, and tochatic analyi with application to traffic engineering, tochatic filtering theory, and mathematical finance.
A note on profit maximization and monotonicity for inbound call centers
A note on profit maximization and monotonicity for inbound call center Ger Koole & Aue Pot Department of Mathematic, Vrije Univeriteit Amterdam, The Netherland 23rd December 2005 Abtract We conider an
More informationGroup Mutual Exclusion Based on Priorities
Group Mutual Excluion Baed on Prioritie Karina M. Cenci Laboratorio de Invetigación en Sitema Ditribuido Univeridad Nacional del Sur Bahía Blanca, Argentina kmc@c.un.edu.ar and Jorge R. Ardenghi Laboratorio
More informationPerformance of Multiple TFRC in Heterogeneous Wireless Networks
Performance of Multiple TFRC in Heterogeneou Wirele Network 1 HyeonJin Jeong, 2 SeongSik Choi 1, Firt Author Computer Engineering Department, Incheon National Univerity, oaihjj@incheon.ac.kr *2,Correponding
More informationStatespace analysis of control systems: Part I
Why a different approach? Statepace analyi of control ytem: Part I Uing a tatevariable approach give u a traightforward way to analyze MIM multipleinput, multiple output ytem. A tate variable model
More informationQueueing systems with scheduled arrivals, i.e., appointment systems, are typical for frontal service systems,
MANAGEMENT SCIENCE Vol. 54, No. 3, March 28, pp. 565 572 in 25199 ein 1526551 8 543 565 inform doi 1.1287/mnc.17.82 28 INFORMS Scheduling Arrival to Queue: A SingleServer Model with NoShow INFORMS
More informationDistributed Monitoring and Aggregation in Wireless Sensor Networks
Ditributed Monitoring and Aggregation in Wirele Senor Network Changlei Liu and Guohong Cao Department of Computer Science & Engineering The Pennylvania State Univerity Email: {chaliu, gcao}@ce.pu.edu
More informationDISTRIBUTED DATA PARALLEL TECHNIQUES FOR CONTENTMATCHING INTRUSION DETECTION SYSTEMS. G. Chapman J. Cleese E. Idle
DISTRIBUTED DATA PARALLEL TECHNIQUES FOR CONTENTMATCHING INTRUSION DETECTION SYSTEMS G. Chapman J. Cleee E. Idle ABSTRACT Content matching i a neceary component of any ignaturebaed network Intruion Detection
More informationDISTRIBUTED DATA PARALLEL TECHNIQUES FOR CONTENTMATCHING INTRUSION DETECTION SYSTEMS
DISTRIBUTED DATA PARALLEL TECHNIQUES FOR CONTENTMATCHING INTRUSION DETECTION SYSTEMS Chritopher V. Kopek Department of Computer Science Wake Foret Univerity WintonSalem, NC, 2709 Email: kopekcv@gmail.com
More informationUtilityBased Flow Control for Sequential Imagery over Wireless Networks
UtilityBaed Flow Control for Sequential Imagery over Wirele Networ Tomer Kihoni, Sara Callaway, and Mar Byer Abtract Wirele enor networ provide a unique et of characteritic that mae them uitable for building
More informationFEDERATION OF ARAB SCIENTIFIC RESEARCH COUNCILS
Aignment Report RP/98983/5/0./03 Etablihment of cientific and technological information ervice for economic and ocial development FOR INTERNAL UE NOT FOR GENERAL DITRIBUTION FEDERATION OF ARAB CIENTIFIC
More informationTwo Dimensional FEM Simulation of Ultrasonic Wave Propagation in Isotropic Solid Media using COMSOL
Excerpt from the Proceeding of the COMSO Conference 0 India Two Dimenional FEM Simulation of Ultraonic Wave Propagation in Iotropic Solid Media uing COMSO Bikah Ghoe *, Krihnan Balaubramaniam *, C V Krihnamurthy
More informationName: SID: Instructions
CS168 Fall 2014 Homework 1 Aigned: Wedneday, 10 September 2014 Due: Monday, 22 September 2014 Name: SID: Dicuion Section (Day/Time): Intruction  Submit thi homework uing Pandagrader/GradeScope(http://www.gradecope.com/
More informationA technical guide to 2014 key stage 2 to key stage 4 value added measures
A technical guide to 2014 key tage 2 to key tage 4 value added meaure CONTENTS Introduction: PAGE NO. What i value added? 2 Change to value added methodology in 2014 4 Interpretation: Interpreting chool
More informationSenior Thesis. Horse Play. Optimal Wagers and the Kelly Criterion. Author: Courtney Kempton. Supervisor: Professor Jim Morrow
Senior Thei Hore Play Optimal Wager and the Kelly Criterion Author: Courtney Kempton Supervior: Profeor Jim Morrow June 7, 20 Introduction The fundamental problem in gambling i to find betting opportunitie
More informationAnalyzing Myopic Approaches for MultiAgent Communication
Analyzing Myopic Approache for MultiAgent Communication Raphen Becker Victor Leer Univerity of Maachuett, Amhert Department of Computer Science {raphen, leer, hlomo}@c.uma.edu Shlomo Zilbertein Abtract
More informationA Duality Model of TCP and Queue Management Algorithms
A Duality Model of TCP and Queue Management Algorithm Steven H. Low CS and EE Department California Intitute of Technology Paadena, CA 95 low@caltech.edu May 4, Abtract We propoe a duality model of endtoend
More informationPartial optimal labeling search for a NPhard subclass of (max,+) problems
Partial optimal labeling earch for a NPhard ubcla of (max,+) problem Ivan Kovtun International Reearch and Training Center of Information Technologie and Sytem, Kiev, Uraine, ovtun@image.iev.ua Dreden
More informationOn the optimality of List Scheduling for online uniform machines scheduling
On the optimality of Lit Scheduling for online uniform machine cheduling Fangqiu HAN Zhiyi TAN Yang YANG Abtract Thi paper conider claical online cheduling problem on uniform machine We how the tight competitive
More informationProfitability of Loyalty Programs in the Presence of Uncertainty in Customers Valuations
Proceeding of the 0 Indutrial Engineering Reearch Conference T. Doolen and E. Van Aken, ed. Profitability of Loyalty Program in the Preence of Uncertainty in Cutomer Valuation Amir Gandomi and Saeed Zolfaghari
More informationA Note on Profit Maximization and Monotonicity for Inbound Call Centers
OPERATIONS RESEARCH Vol. 59, No. 5, September October 2011, pp. 1304 1308 in 0030364X ein 15265463 11 5905 1304 http://dx.doi.org/10.1287/opre.1110.0990 2011 INFORMS TECHNICAL NOTE INFORMS hold copyright
More informationAvailability of WDM Multi Ring Networks
Paper Availability of WDM Multi Ring Network Ivan Rado and Katarina Rado H d.o.o. Motar, Motar, Bonia and Herzegovina Faculty of Electrical Engineering, Mechanical Engineering and Naval Architecture, Univerity
More informationDUE to the small size and low cost of a sensor node, a
1992 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 14, NO. 10, OCTOBER 2015 A Networ Coding Baed Energy Efficient Data Bacup in SurvivabilityHeterogeneou Senor Networ Jie Tian, Tan Yan, and Guiling Wang
More informationSCM integration: organiational, managerial and technological iue M. Caridi 1 and A. Sianei 2 Dipartimento di Economia e Produzione, Politecnico di Milano, Italy Email: maria.caridi@polimi.it Itituto
More informationMobile Network Configuration for Largescale Multimedia Delivery on a Single WLAN
Mobile Network Configuration for Largecale Multimedia Delivery on a Single WLAN Huigwang Je, Dongwoo Kwon, Hyeonwoo Kim, and Hongtaek Ju Dept. of Computer Engineering Keimyung Univerity Daegu, Republic
More informationMixed Method of Model Reduction for Uncertain Systems
SERBIAN JOURNAL OF ELECTRICAL ENGINEERING Vol 4 No June Mixed Method of Model Reduction for Uncertain Sytem N Selvaganean Abtract: A mixed method for reducing a higher order uncertain ytem to a table reduced
More informationAssessing the Discriminatory Power of Credit Scores
Aeing the Dicriminatory Power of Credit Score Holger Kraft 1, Gerald Kroiandt 1, Marlene Müller 1,2 1 Fraunhofer Intitut für Techno und Wirtchaftmathematik (ITWM) GottliebDaimlerStr. 49, 67663 Kaierlautern,
More informationA Resolution Approach to a Hierarchical Multiobjective Routing Model for MPLS Networks
A Reolution Approach to a Hierarchical Multiobjective Routing Model for MPLS Networ Joé Craveirinha a,c, Rita GirãoSilva a,c, João Clímaco b,c, Lúcia Martin a,c a b c DEECFCTUC FEUC INESCCoimbra International
More information1 Safe Drivers versus Reckless Drunk Drivers
ECON 301: General Equilibrium IV (Externalitie) 1 Intermediate Microeconomic II, ECON 301 General Equilibrium IV: Externalitie In our dicuion thu far, we have implicitly aumed that all good can be traded
More informationOptical Illusion. Sara Bolouki, Roger Grosse, Honglak Lee, Andrew Ng
Optical Illuion Sara Bolouki, Roger Groe, Honglak Lee, Andrew Ng. Introduction The goal of thi proect i to explain ome of the illuory phenomena uing pare coding and whitening model. Intead of the pare
More informationClusterAware Cache for Network Attached Storage *
CluterAware Cache for Network Attached Storage * Bin Cai, Changheng Xie, and Qiang Cao National Storage Sytem Laboratory, Department of Computer Science, Huazhong Univerity of Science and Technology,
More informationNETWORK TRAFFIC ENGINEERING WITH VARIED LEVELS OF PROTECTION IN THE NEXT GENERATION INTERNET
Chapter 1 NETWORK TRAFFIC ENGINEERING WITH VARIED LEVELS OF PROTECTION IN THE NEXT GENERATION INTERNET S. Srivatava Univerity of Miouri Kana City, USA hekhar@conrel.ice.umkc.edu S. R. Thirumalaetty now
More informationSolutions to Sample Problems for Test 3
22 Differential Equation Intructor: Petronela Radu November 8 25 Solution to Sample Problem for Tet 3 For each of the linear ytem below find an interval in which the general olution i defined (a) x = x
More informationSocially Optimal Pricing of Cloud Computing Resources
Socially Optimal Pricing of Cloud Computing Reource Ihai Menache Microoft Reearch New England Cambridge, MA 02142 timena@microoft.com Auman Ozdaglar Laboratory for Information and Deciion Sytem Maachuett
More informationDesign of Compound Hyperchaotic System with Application in Secure Data Transmission Systems
Deign of Compound Hyperchaotic Sytem with Application in Secure Data Tranmiion Sytem D. Chantov Key Word. Lyapunov exponent; hyperchaotic ytem; chaotic ynchronization; chaotic witching. Abtract. In thi
More informationLecture Notes for Laplace Transform
Lecture Note for Laplace Tranform Wen Shen April 9 NB! Thee note are ued by myelf. They are provided to tudent a a upplement to the textbook. They can not ubtitute the textbook. Laplace Tranform i ued
More informationScheduling of Jobs and Maintenance Activities on Parallel Machines
Scheduling of Job and Maintenance Activitie on Parallel Machine ChungYee Lee* Department of Indutrial Engineering Texa A&M Univerity College Station, TX 778433131 cylee@ac.tamu.edu ZhiLong Chen** Department
More informationA New Optimum Jitter Protection for Conversational VoIP
Proc. Int. Conf. Wirele Commun., Signal Proceing (Nanjing, China), 5 pp., Nov. 2009 A New Optimum Jitter Protection for Converational VoIP Qipeng Gong, Peter Kabal Electrical & Computer Engineering, McGill
More informationHarmonic Oscillations / Complex Numbers
Harmonic Ocillation / Complex Number Overview and Motivation: Probably the ingle mot important problem in all of phyic i the imple harmonic ocillator. It can be tudied claically or uantum mechanically,
More informationMECH 2110  Statics & Dynamics
Chapter D Problem 3 Solution 1/7/8 1:8 PM MECH 11  Static & Dynamic Chapter D Problem 3 Solution Page 7, Engineering Mechanic  Dynamic, 4th Edition, Meriam and Kraige Given: Particle moving along a traight
More informationChapter 4: MeanVariance Analysis
Chapter 4: MeanVariance Analyi Modern portfolio theory identifie two apect of the invetment problem. Firt, an invetor will want to maximize the expected rate of return on the portfolio. Second, an invetor
More information1 Introduction. Reza Shokri* Privacy Games: Optimal UserCentric Data Obfuscation
Proceeding on Privacy Enhancing Technologie 2015; 2015 (2):1 17 Reza Shokri* Privacy Game: Optimal UerCentric Data Obfucation Abtract: Conider uer who hare their data (e.g., location) with an untruted
More informationUnit 11 Using Linear Regression to Describe Relationships
Unit 11 Uing Linear Regreion to Decribe Relationhip Objective: To obtain and interpret the lope and intercept of the leat quare line for predicting a quantitative repone variable from a quantitative explanatory
More informationOnline story scheduling in web advertising
Online tory cheduling in web advertiing Anirban Dagupta Arpita Ghoh Hamid Nazerzadeh Prabhakar Raghavan Abtract We tudy an online job cheduling problem motivated by toryboarding in web advertiing, where
More informationv = x t = x 2 x 1 t 2 t 1 The average speed of the particle is absolute value of the average velocity and is given Distance travelled t
Chapter 2 Motion in One Dimenion 2.1 The Important Stuff 2.1.1 Poition, Time and Diplacement We begin our tudy of motion by conidering object which are very mall in comparion to the ize of their movement
More informationAlgorithms for Advance Bandwidth Reservation in Media Production Networks
Algorithm for Advance Bandwidth Reervation in Media Production Network Maryam Barhan 1, Hendrik Moen 1, Jeroen Famaey 2, Filip De Turck 1 1 Department of Information Technology, Ghent Univerity imind Gaton
More informationINFORMATION Technology (IT) infrastructure management
IEEE TRANSACTIONS ON CLOUD COMPUTING, VOL. 2, NO. 1, MAY 214 1 BuineDriven Longterm Capacity Planning for SaaS Application David Candeia, Ricardo Araújo Santo and Raquel Lope Abtract Capacity Planning
More informationHow to Maximize User Satisfaction Degree in Multiservice IP Networks
How to Maximize Uer Satifaction Degree in Multiervice IP Network Huy Anh Nguyen, Tam Van Nguyen and Deokai Choi Department of Electronic and Computer Engineering Chonnam National Univerity Gwangu, KOREA
More informationMODIFIED 2D FINITEDIFFERENCE TIMEDOMAIN TECHNIQUE FOR TUNNEL PATH LOSS PREDICTION. Y. Wu, M. Lin and I.J. Wassell
2 nd International Conference on Wirele Communication in Underground and Confined Area Augut 2527, 2008 Vald Or  Québec  Canada MODIFIED 2D FINITEDIFFERENCE TIMEDOMAIN TECHNIQUE FOR TUNNEL PATH LOSS
More informationPerformance of a BrowserBased JavaScript Bandwidth Test
Performance of a BrowerBaed JavaScript Bandwidth Tet David A. Cohen II May 7, 2013 CP SC 491/H495 Abtract An exiting browerbaed bandwidth tet written in JavaScript wa modified for the purpoe of further
More informationCHARACTERISTICS OF WAITING LINE MODELS THE INDICATORS OF THE CUSTOMER FLOW MANAGEMENT SYSTEMS EFFICIENCY
Annale Univeritati Apuleni Serie Oeconomica, 2(2), 200 CHARACTERISTICS OF WAITING LINE MODELS THE INDICATORS OF THE CUSTOMER FLOW MANAGEMENT SYSTEMS EFFICIENCY Sidonia Otilia Cernea Mihaela Jaradat 2 Mohammad
More informationProject Management Basics
Project Management Baic A Guide to undertanding the baic component of effective project management and the key to ucce 1 Content 1.0 Who hould read thi Guide... 3 1.1 Overview... 3 1.2 Project Management
More informationINTERACTIVE TOOL FOR ANALYSIS OF TIMEDELAY SYSTEMS WITH DEADTIME COMPENSATORS
INTERACTIVE TOOL FOR ANALYSIS OF TIMEDELAY SYSTEMS WITH DEADTIME COMPENSATORS Joé Lui Guzmán, Pedro García, Tore Hägglund, Sebatián Dormido, Pedro Alberto, Manuel Berenguel Dep. de Lenguaje y Computación,
More informationDistributed, Secure Load Balancing with Skew, Heterogeneity, and Churn
Ditributed, Secure Load Balancing with Skew, Heterogeneity, and Churn Jonathan Ledlie and Margo Seltzer Diviion of Engineering and Applied Science Harvard Univerity Abtract Numerou propoal exit for load
More informationJanuary 21, 2015. Abstract
T S U I I E P : T R M C S J. R January 21, 2015 Abtract Thi paper evaluate the trategic behavior of a monopolit to influence environmental policy, either with taxe or with tandard, comparing two alternative
More informationQueueing Models for Multiclass Call Centers with RealTime Anticipated Delays
Queueing Model for Multicla Call Center with RealTime Anticipated Delay Oualid Jouini Yve Dallery Zeynep Akşin Ecole Centrale Pari Koç Univerity Laboratoire Génie Indutriel College of Adminitrative Science
More informationReview of Multiple Regression Richard Williams, University of Notre Dame, http://www3.nd.edu/~rwilliam/ Last revised January 13, 2015
Review of Multiple Regreion Richard William, Univerity of Notre Dame, http://www3.nd.edu/~rwilliam/ Lat revied January 13, 015 Aumption about prior nowledge. Thi handout attempt to ummarize and yntheize
More informationTRADING rules are widely used in financial market as
Complex Stock Trading Strategy Baed on Particle Swarm Optimization Fei Wang, Philip L.H. Yu and David W. Cheung Abtract Trading rule have been utilized in the tock market to make profit for more than a
More informationGrowth and Sustainability of Managed Security Services Networks: An Economic Perspective
Growth and Sutainability of Managed Security Service etwork: An Economic Perpective Alok Gupta Dmitry Zhdanov Department of Information and Deciion Science Univerity of Minneota Minneapoli, M 55455 (agupta,
More informationHeuristic Approach to Dynamic Data Allocation in Distributed Database Systems
Pakitan Journal of Information and Technology 2 (3): 231239, 2003 ISSN 16826027 2003 Aian Network for Scientific Information Heuritic Approach to Dynamic Data Allocation in Ditributed Databae Sytem 1
More informationAuctionBased Resource Allocation for Sharing Cloudlets in Mobile Cloud Computing
1 AuctionBaed Reource Allocation for Sharing Cloudlet in Mobile Cloud Computing ALong Jin, Wei Song, Senior Member, IEEE, and Weihua Zhuang, Fellow, IEEE Abtract Driven by pervaive mobile device and
More informationRedesigning Ratings: Assessing the Discriminatory Power of Credit Scores under Censoring
Redeigning Rating: Aeing the Dicriminatory Power of Credit Score under Cenoring Holger Kraft, Gerald Kroiandt, Marlene Müller Fraunhofer Intitut für Techno und Wirtchaftmathematik (ITWM) Thi verion: June
More informationMSc Financial Economics: International Finance. Bubbles in the Foreign Exchange Market. Anne Sibert. Revised Spring 2013. Contents
MSc Financial Economic: International Finance Bubble in the Foreign Exchange Market Anne Sibert Revied Spring 203 Content Introduction................................................. 2 The Mone Market.............................................
More informationBidding for Representative Allocations for Display Advertising
Bidding for Repreentative Allocation for Diplay Advertiing Arpita Ghoh, Preton McAfee, Kihore Papineni, and Sergei Vailvitkii Yahoo! Reearch. {arpita, mcafee, kpapi, ergei}@yahooinc.com Abtract. Diplay
More informationHealth Insurance and Social Welfare. Run Liang. China Center for Economic Research, Peking University, Beijing 100871, China,
Health Inurance and Social Welfare Run Liang China Center for Economic Reearch, Peking Univerity, Beijing 100871, China, Email: rliang@ccer.edu.cn and Hao Wang China Center for Economic Reearch, Peking
More informationAnalysis of Mesostructure Unit Cells Comprised of Octettruss Structures
Analyi of Meotructure Unit Cell Compried of Octettru Structure Scott R. Johnton *, Marque Reed *, Hongqing V. Wang, and David W. Roen * * The George W. Woodruff School of Mechanical Engineering, Georgia
More informationIn this paper, we investigate toll setting as a policy tool to regulate the use of roads for dangerous goods
Vol. 43, No. 2, May 2009, pp. 228 243 in 00411655 ein 15265447 09 4302 0228 inform doi 10.1287/trc.1080.0236 2009 INFORMS Toll Policie for Mitigating Hazardou Material Tranport Rik Patrice Marcotte,
More informationLinear energypreserving integrators for Poisson systems
BIT manucript No. (will be inerted by the editor Linear energypreerving integrator for Poion ytem David Cohen Ernt Hairer Received: date / Accepted: date Abtract For Hamiltonian ytem with noncanonical
More informationBiObjective Optimization for the Clinical Trial Supply Chain Management
Ian David Lockhart Bogle and Michael Fairweather (Editor), Proceeding of the 22nd European Sympoium on Computer Aided Proce Engineering, 1720 June 2012, London. 2012 Elevier B.V. All right reerved. BiObjective
More informationRegulated Maximal Matching: A Distributed Scheduling Algorithm for MultiHop Wireless Networks With NodeExclusive Spectrum Sharing
Proceeding of the 44th IEEE Conference on Deciion and Control, and the Euroean Control Conference 005 Seville, Sain, December 115, 005 WeB14.5 Regulated Maximal Matching: A Ditributed Scheduling Algorithm
More informationGrowth and Sustainability of Managed Security Services Networks: An Economic Perspective
Growth and Sutainability of Managed Security Service etwork: An Economic Perpective Alok Gupta Dmitry Zhdanov Department of Information and Deciion Science Univerity of Minneota Minneapoli, M 55455 (agupta,
More informationAuction Mechanisms Toward Efficient Resource Sharing for Cloudlets in Mobile Cloud Computing
1 Auction Mechanim Toward Efficient Reource Sharing for Cloudlet in Mobile Cloud Computing ALong Jin, Wei Song, Ping Wang, Duit Niyato, and Peijian Ju Abtract Mobile cloud computing offer an appealing
More informationCHAPTER 5 BROADBAND CLASSE AMPLIFIER
CHAPTER 5 BROADBAND CLASSE AMPLIFIER 5.0 Introduction ClaE amplifier wa firt preented by Sokal in 1975. The application of cla E amplifier were limited to the VHF band. At thi range of frequency, clae
More informationChapter 10 Stocks and Their Valuation ANSWERS TO ENDOFCHAPTER QUESTIONS
Chapter Stoc and Their Valuation ANSWERS TO ENOFCHAPTER QUESTIONS  a. A proxy i a document giving one peron the authority to act for another, typically the power to vote hare of common toc. If earning
More informationCASE STUDY ALLOCATE SOFTWARE
CASE STUDY ALLOCATE SOFTWARE allocate caetud y TABLE OF CONTENTS #1 ABOUT THE CLIENT #2 OUR ROLE #3 EFFECTS OF OUR COOPERATION #4 BUSINESS PROBLEM THAT WE SOLVED #5 CHALLENGES #6 WORKING IN SCRUM #7 WHAT
More informationHUMAN CAPITAL AND THE FUTURE OF TRANSITION ECONOMIES * Michael Spagat Royal Holloway, University of London, CEPR and Davidson Institute.
HUMAN CAPITAL AND THE FUTURE OF TRANSITION ECONOMIES * By Michael Spagat Royal Holloway, Univerity of London, CEPR and Davidon Intitute Abtract Tranition economie have an initial condition of high human
More informationTransmission Scheduling for MultiChannel Satellite and Wireless Networks
Tranmiion Scheduling for ultichannel Satellite and Wirele Network Anand Ganti, Eytan odiano, John N. Titikli aachuett Intitute of Technology email: aganti@mit.edu, modiano@mit.edu, jnt@mit.edu Abtract
More informationNetwork Architecture for Joint Failure Recovery and Traffic Engineering
Network Architecture for Joint Failure Recovery and Traffic Engineering Martin Suchara Dept. of Computer Science Princeton Univerity, NJ 08544 muchara@princeton.edu Dahai Xu AT&T Lab Reearch Florham Park,
More informationRisk Management for a Global Supply Chain Planning under Uncertainty: Models and Algorithms
Rik Management for a Global Supply Chain Planning under Uncertainty: Model and Algorithm Fengqi You 1, John M. Waick 2, Ignacio E. Gromann 1* 1 Dept. of Chemical Engineering, Carnegie Mellon Univerity,
More informationREDUCTION OF TOTAL SUPPLY CHAIN CYCLE TIME IN INTERNAL BUSINESS PROCESS OF REAMER USING DOE AND TAGUCHI METHODOLOGY. Abstract. 1.
International Journal of Advanced Technology & Engineering Reearch (IJATER) REDUCTION OF TOTAL SUPPLY CHAIN CYCLE TIME IN INTERNAL BUSINESS PROCESS OF REAMER USING DOE AND Abtract TAGUCHI METHODOLOGY Mr.
More informationA model for the relationship between tropical precipitation and column water vapor
Click Here for Full Article GEOPHYSICAL RESEARCH LETTERS, VOL. 36, L16804, doi:10.1029/2009gl039667, 2009 A model for the relationhip between tropical precipitation and column water vapor Caroline J. Muller,
More informationChapter and. FIGURE 9 36 The deviation of an actual gasturbine cycle from the ideal Brayton cycle as a result of irreversibilities.
Chapter 9 The thermal efficiency could alo be determined from where h th q out q out h h 789.7 00.9 89. kj>kg Dicuion Under the coldairtard aumption (contant pecific heat value at room temperature),
More informationUsing Graph Analysis to Study Networks of Adaptive Agent
Uing Graph Analyi to Study Network of Adaptive Agent Sherief Abdallah Britih Univerity in Dubai, United Arab Emirate Univerity of Edinburgh, United Kingdom hario@ieee.org ABSTRACT Experimental analyi of
More informationControl Theory based Approach for the Improvement of Integrated Business Process Interoperability
www.ijcsi.org 201 Control Theory baed Approach for the Improvement of Integrated Buine Proce Interoperability Abderrahim Taoudi 1, Bouchaib Bounabat 2 and Badr Elmir 3 1 AlQualadi Reearch & Development
More informationMobility Improves Coverage of Sensor Networks
Mobility Improve Coverage of Senor Networ Benyuan Liu Dept. of Computer Science Univerity of MaachuettLowell Lowell, MA 1854 Peter Bra Dept. of Computer Science City College of New Yor New Yor, NY 131
More informationDecision Theory and the Normal Distribution
MODULE 3 Deciion Theory and the Normal Ditribution LEARNING OBJECTIVE After completing thi module, tudent will be able to: 1. Undertand how the normal curve can be ued in performing breakeven analyi.
More informationGabriel E. Arrobo and Richard D. Gitlin, NAI Charter Fellow
Technology and Innovation, Vol. 15, pp. 227 236, 2013 19498241/13 $90.00 +.00 Printed in the USA. All right reerved. DOI: http://dx.doi.org/10.3727/194982413x13790020921825 Copyright ã 2013 Cognizant
More informationResearch Article An (s, S) Production Inventory Controlled SelfService Queuing System
Probability and Statitic Volume 5, Article ID 558, 8 page http://dxdoiorg/55/5/558 Reearch Article An (, S) Production Inventory Controlled SelfService Queuing Sytem Anoop N Nair and M J Jacob Department
More informationAccelerationDisplacement Crash Pulse Optimisation A New Methodology to Optimise Vehicle Response for Multiple Impact Speeds
AccelerationDiplacement Crah Pule Optimiation A New Methodology to Optimie Vehicle Repone for Multiple Impact Speed D. Gildfind 1 and D. Ree 2 1 RMIT Univerity, Department of Aeropace Engineering 2 Holden
More informationIntroduction to DoD Architecture Framework (DoDAF)
Introduction to DoD Architecture Framework (DoDAF) Beverly Hawk 1 Jonnathan Kim 2 What i thi paper? Thi paper, Introduction to the DoD Architecture Framework (DoDAF), i an ultrabrief overview of DoDAF.
More informationApigee Edge: Apigee Cloud vs. Private Cloud. Evaluating deployment models for API management
Apigee Edge: Apigee Cloud v. Private Cloud Evaluating deployment model for API management Table of Content Introduction 1 Time to ucce 2 Total cot of ownerhip 2 Performance 3 Security 4 Data privacy 4
More informationCASE STUDY BRIDGE. www.futureprocessing.com
CASE STUDY BRIDGE TABLE OF CONTENTS #1 ABOUT THE CLIENT 3 #2 ABOUT THE PROJECT 4 #3 OUR ROLE 5 #4 RESULT OF OUR COLLABORATION 67 #5 THE BUSINESS PROBLEM THAT WE SOLVED 8 #6 CHALLENGES 9 #7 VISUAL IDENTIFICATION
More informationReturn on Investment and Effort Expenditure in the Software Development Environment
International Journal of Applied Information ytem (IJAI) IN : 22490868 Return on Invetment and Effort Expenditure in the oftware Development Environment Dineh Kumar aini Faculty of Computing and IT, ohar
More informationReport 46681b 30.10.2010. Measurement report. Sylomer  field test
Report 46681b Meaurement report Sylomer  field tet Report 46681b 2(16) Contet 1 Introduction... 3 1.1 Cutomer... 3 1.2 The ite and purpoe of the meaurement... 3 2 Meaurement... 6 2.1 Attenuation of
More informationChapter 8 Transients & RLC
Chapter 8 Tranient & C Chapter 8 Tranient & C... 8. Introduction... 8.3 Tranient... 8.3. Firt Order Tranient... 8.3. Circuit... 3 8.3.3 C Circuit... 4 8.9 C Sytem epone... 5 8.9. C Equation... 5 8.9. Sytem
More informationExposure Metering Relating Subject Lighting to Film Exposure
Expoure Metering Relating Subject Lighting to Film Expoure By Jeff Conrad A photographic expoure meter meaure ubject lighting and indicate camera etting that nominally reult in the bet expoure of the film.
More informationGrowing SelfOrganizing Maps for Surface Reconstruction from Unstructured Point Clouds
Growing SelfOrganizing Map for Surface Recontruction from Untructured Point Cloud Renata L. M. E. do Rêgo, Aluizio F. R. Araújo, and Fernando B.de Lima Neto Abtract Thi work introduce a new method for
More informationOn a Generalization of Meyniel s Conjecture on the Cops and Robbers Game
On a Generalization of Meyniel Conjecture on the Cop and Robber Game Noga Alon Tel Aviv Univerity and Intitute for Advanced Study, Princeton nogaa@tau.ac.il Abba Mehrabian Department of Combinatoric and
More informationAMS577. Repeated Measures ANOVA: The Univariate and the Multivariate Analysis Approaches
AMS577. Repeated Meaure ANOVA: The Univariate and the Multivariate Analyi Approache 1. Oneway Repeated Meaure ANOVA Oneway (onefactor) repeatedmeaure ANOVA i an extenion of the matchedpair ttet to
More informationSELFMANAGING PERFORMANCE IN APPLICATION SERVERS MODELLING AND DATA ARCHITECTURE
SELFMANAGING PERFORMANCE IN APPLICATION SERVERS MODELLING AND DATA ARCHITECTURE RAVI KUMAR G 1, C.MUTHUSAMY 2 & A.VINAYA BABU 3 1 HP Bangalore, Reearch Scholar JNTUH, Hyderabad, India, 2 Yahoo, Bangalore,
More informationTIME SERIES ANALYSIS AND TRENDS BY USING SPSS PROGRAMME
TIME SERIES ANALYSIS AND TRENDS BY USING SPSS PROGRAMME RADMILA KOCURKOVÁ Sileian Univerity in Opava School of Buine Adminitration in Karviná Department of Mathematical Method in Economic Czech Republic
More information