Pricing Network Resources for Adaptive Applications in a Differentiated Services Network

Size: px
Start display at page:

Download "Pricing Network Resources for Adaptive Applications in a Differentiated Services Network"

Transcription

1 IEEE INFOCOM Prcng Network Resources for Adaptve Applcatons n a Dfferentated Servces Network Xn Wang and Hennng Schulzrnne Columba Unversty Emal: {xnwang, schulzrnne}@cs.columba.edu Abstract The Dfferentated Servces framework (DffServ) [] has been proposed to provde multple Qualty of Servce (QoS) classes over IP networks. A network supportng multple classes of servce also requres a dfferentated prcng structure. In ths work, we propose a prcng scheme n a DffServ envronment based on the cost of provdng dfferent levels of qualty of servce to dfferent classes, and on long-term demand. Prcng of network servces dynamcally based on the level of servce, usage, and congeston allows a more compettve prce to be offered, allows the network to be used more effcently, and provdes a natural and equtable ncentve for applcatons to adapt ther servce contract accordng to network condtons. We develop a DffServ smulaton framework to compare the performance of a network supportng congeston-senstve prcng and adaptve servce negotaton to that of a network wth a statc prcng polcy. Adaptve users adapt to prce changes by adjustng ther sendng rate or selectng a dfferent servce class. We also develop the demand behavor of adaptve users based on a physcally reasonable user utlty functon. Smulaton results show that a congeston-senstve prcng polcy coupled wth user rate adaptaton s able to control congeston and allow a servce class to meet ts performance assurances under large or bursty offered loads, even wthout explct admsson control. Users are able to mantan a stable expendture. Allowng users to mgrate between servce classes n response to prce ncreases further stablzes the ndvdual servce prces. When admsson control s enforced, congeston-senstve prcng stll provdes an advantage n terms of a much lower connecton blockng rate at hgh loads. I. INTRODUCTION The Dfferentated Servces framework (DffServ) [] has been proposed to provde multple Qualty of Servce (QoS) classes over IP networks. Two types of Per-Hop-Behavor (PHB) are proposed: Expedted Forwardng (EF) [] and Assured Forwardng (AF) [3]. The EF PHB s defned as a forwardng treatment where the departure rate of an aggregate s packets from any DffServ node must equal or exceed a confgurable rate. For AF servce, four classes wth three levels of drop precedence n each class are defned for general use. A network supportng multple classes of servce also requres a dfferentated prcng structure, rather than the flatfee prcng model adopted by vrtually all current Internet servces. Whle network tarff structures are often domnated by busness and marketng arguments rather than costs, we beleve t s worthwhle to understand and develop a cost-based prcng structure as a gude for actual prcng. In economcally Ths work was sponsored by NSF CAREER grant. vable models, the dfference n the charge between dfferent servce classes would presumably depend on the dfference n performance between the classes, and should take nto account the average (long-term) demand for each class. In general, the level of forwardng assurance of an IP packet n DffServ depends on the amount of resources allocated to a class the packet belongs to, the current load of the class, and n case of congeston wthn the class, the drop precedence of the packet. Also, when multple servces are avalable at dfferent prces, users should be able to demand partcular servces, sgnal the network to provson accordng to the requested qualty, and generate accountng and bllng records. One of the two man goals of our work s to develop a prcng scheme n a DffServ envronment based on the cost of provdng dfferent levels of qualty of servce to dfferent classes, and on long-term demand. DffServ supports servces whch nvolve a traffc contract or servce level agreement (SLA) between the user and the network. If the agreement, ncludng prce negotaton and resource allocaton are set statcally (before transmsson), prcng, resource allocaton and admsson control polces (f any) have to be conservatve to be able to meet QoS assurances n the presence of network traffc dynamcs. Prcng of network servces dynamcally based on the level of servce, usage, and congeston allows a more compettve prce to be offered, and allows the network to be used more effcently. Dfferentated and congeston-senstve prcng also provdes a natural and equtable ncentve for applcatons to adapt ther servce contract accordng to network condtons. A number of adaptaton schemes have been proposed for multmeda applcatons to dynamcally regulate the source bandwdth accordng to the exstng network condtons (a survey of ths work s gven n [4]). The second man goal of our work s to ntegrate our prcng scheme wth a dynamc prcng and servce negotaton envronment. In ths envronment, servce prces have a congestonsenstve component n addton to the long-term, relatvely statc prce. Some or all users are adaptaton-capable, and adapt to prce changes by adjustng ther sendng rate or selectng a dfferent servce class. Users wth strngent bandwdth and QoS requrements mantan a hgh qualty by payng more, whle adaptaton-ncapable applcatons use servces offerng a statc prce. We develop the demand behavor of adaptve users based on a physcally reasonable user utlty functon. In our smulatons, prces and servces are negotated through a Resource Negotaton and Prcng (RNAP) protocol

2 IEEE INFOCOM and archtecture, presented n earler work [5]. RNAP enables the user to select from avalable network servces wth dfferent QoS propertes and re-negotate contracted servces, and enables the network to dynamcally formulate servce prces and communcate current prces to the user. In RNAP, resource commtments are typcally made for short negotaton ntervals, nstead of ndefntely, and prces may vary for each nterval. Usng RNAP and an extended verson of an exstng Dff- Serv mplementaton, we develop a smulaton framework to compare the performance of a network supportng congestonsenstve prcng and adaptve servce negotaton to that of a network wth a statc prcng polcy. We also study the stablty of the dynamc prcng and servce negotaton mechansms. We evaluate the system performance and perceved beneft (or value-for-money) under the dynamc and statc systems. We also study the relatve effects on system performance of rate adaptaton, dynamc load balancng between servce classes and admsson. Although the smulaton framework s based on the RNAP model, we try to derve results and conclusons applcable to statc and congeston-drven, dynamc prcng schemes n general. Ths paper s organzed as follows. Secton II develops a physcally realstc user utlty functon to represent user demand behavor n response to prce changes. Secton III dscusses our proposed prcng model n detal. Secton IV summarzes our earler work on RNAP and partcularly how t supports network prcng. In secton V we descrbe our smulaton model, and n secton VI we dscuss smulaton results. We descrbe some related work n secton VII, and summarze our work n secton VIII. II. USER ADAPTATION In a network wth congeston dependent prcng and dynamc resource negotaton (through RNAP or some other sgnalng protocol), adaptve applcatons wth a budget constrant wll adjust ther servce requests n response to prce varatons. In ths secton, we dscuss how a set of user applcatons performng a gven task (for example, a vdeo conference) adapt ther sendng rate and qualty of servce requests to the network n response to changes n servce prces, so as to maxmze the beneft or utlty to the user, subject to the constrant of the user s budget. Although we focus on adaptve applcatons as the ones best suted to a dynamc prcng envronment, the RNAP framework does not requre adaptaton capablty. Applcatons may choose servces that provde a fxed prce and fxed servce parameters durng the duraton of servce. Generally, the longterm average cost for a fxed-prce servce wll be hgher, snce t uses network resources less effcently. Alternatvely, applcatons may use a servce wth usage-senstve prcng, and mantan a hgh QoS level, payng a hgher charge durng congeston. We consder a set of user applcatons, requred to perform ataskormsson. The user would lke to determne a set of transmsson parameters (sendng rate and QoS parameters) from whch t can derve the maxmum beneft, subject to hs budget. We assume that the user defnes quanttatvely, through a utlty functon, the perceved monetary value (say, 5 cents/mnute) provded by the set of transmsson parameters towards completng the msson. Consumers n the real world generally try to obtan the best possble value for the money they pay, subject to ther budget and mnmum qualty requrements; n other words, consumers may prefer lower qualty at a lower prce f they perceve ths as meetng ther requrements and offerng better value. Intutvely, ths seems to be a reasonable model n a network wth QoS support, where the user pays for the level of QoS he receves. In our case, the value for money obtaned by the user corresponds to the surplus between the utlty U( ) wth a partcular set of transmsson parameters (snce ths s the perceved value), and the cost of obtanng that servce. The goal of the adaptaton s to maxmze ths surplus, subject to the budget and the mnmum and maxmum QoS requrements. We now consder the smultaneous adaptaton of transmsson parameters of a set of n applcatons performng a sngle task. The transmsson bandwdth and QoS parameters for each applcaton are selected and adapted so as to maxmze the msson-wde value perceved by the user, as represented by the surplus of the total utlty, Û, over the total cost C. We can thnk of the adaptaton process as the allocaton and dynamc re-allocaton of a fnte amount of resources between the applcatons. In ths paper, we make the smplfyng assumpton that for each applcaton, a utlty functon can be defned as a functon only of the transmsson parameters of that applcaton, ndependent of the transmsson parameters of other applcatons. Snce we consder utlty to be equvalent to a certan monetary value, we can wrte the total utlty as the sum of ndvdual applcaton utltes : Û = [U (x (T spec,r spec)] () where x s the transmsson (T sepc ) and qualty of servce parameter (R spec ) tuple for the th applcaton. The optmzaton of surplus can be wrtten as max [U (x ) C (x )] s. t. C (x ) b, x mn x x max () where x mn and x max represent the mnmum and maxmum transmsson requrements for stream, C s the cost of the type of servce selected for stream at requested transmsson parameter x,andbs the budget of the user. In practce, the applcaton utlty s lkely to be measured by user experments and known at dscrete bandwdths, at one or a few levels of loss and delay, possbly correspondng to a subset of the avalable servces; at the current stage of research, some possble servces are guaranteed [6] and controlled-load servce [7] under the nt-serv model, Expedted Forwardng (EF) [] and Assured Forwardng (AF) [3] under dff-serv. In ths case, t s convenent to represent the utlty as a pecewse lnear functon of bandwdth (or a set of such functons). A smplfed algorthm s proposed n [8] to search for the optmal servce requests n such a framework.

3 IEEE INFOCOM 3 We can make some general assumptons about the utlty functon as a functon of the bandwdth (can be equvalent bandwdth [9]), at a fxed value of loss and delay. A user applcaton generally has a mnmum bandwdth requrement. It also assocates a certan mnmum value wth a task, whch may be regarded as an opportunty value, and ths s the perceved utlty when the applcaton receves just the mnmum requred bandwdth. The user termnates the applcaton f ts mnmum bandwdth requrement can not be fulflled, or when the prce charged s hgher than the opportunty value derved from keepng the connecton alve. Also, user experments reported n the lterature [][] suggest that utlty functons typcally follow a model of dmnshng returns to scale, that s, the margnal utlty as a functon of bandwdth dmnshes wth ncreasng bandwdth. Hence, a utlty functon can be represented n a general form as a functon of bandwdth as: U(x) =U + w log x (3) x m where x m represents the mnmum bandwdth the applcaton requres, w represents the senstvty of the utlty to bandwdth, and U s the monetary opportunty that the user perceves at the lowest QoS level. The utlty functon s also senstve to network transmsson parameters such as loss and delay. In our work, we rely on the expermental results n [] whch show that users perceved qualty for nteractve audo decreases almost lnearly wth ether delay or loss, wth a mnmum acceptable qualty requrement. More subjectve tests are needed for other applcaton types. Currently, we assume a smlar lnear dependence for all applcatons. Accordngly, we represent the utlty functon as: U(x) =U + w log x k d d k l l, for x x m, (4) x m where k d and k l represent respectvely the user s senstvty to delay and loss. In some cases, the user s perceved senstvty may depend on the bandwdth used. For example, tolerance to delay and loss wll be dfferent for dfferent speech codecs. Snce we are not assumng any partcular applcaton model, we assume users delay and loss senstvty are bandwdth ndependent n our smulatons. A user wth a hgher senstvty to delay or loss wll tend to select a hgher servce class rather than request more bandwdth. If the utltes of all the applcatons are represented n the format of equaton 4, the optmzaton process for a system wth multple applcatons can be represented as: max [U + w log x kdd kll p x ] x m s. t. p x b, x x m,, d D, l L (5) where p s the prce of the servce class selected by the applcaton, D and L are respectvely the loss and delay bound of an applcaton, above whch the applcaton no longer functons usefully. It s possble to represent the above optmzaton problem as a Lagrangan and solve t. However, we assume the avalablty of only a few dfferent loss and delay levels correspondng to dfferent servce classes, and accordngly use a more heurstc method. The optmzaton nvolves assgnng a servce class and a bandwdth to each applcaton. For a partcular assgnment of servce classes to applcatons, f the user can obtan the optmal bandwdth dstrbuton accordng to equaton 5 at a cost below hs budget, then the bandwdth allocaton that maxmzes the perceved surplus for an applcaton can be showntobe: x = w p (6) Hence, w represents the money a user would spend based on ts perceved value for an applcaton.the above bandwdth dstrbuton s consdered for all possble servce class assgnments (constraned by applcaton requrements and budget), and the one gvng the hghest total surplus s used. If there s no set of servce class assgnments for whch the optmal dstrbuton of equaton 6 can be obtaned at a cost below the budget, the total budget s frst dstrbuted to the component applcatons accordng to ther relatve bandwdth senstvty w. That s, each applcaton receves a budget share b such that b = b w (7) k wk Each applcaton s then allocated a servce and bandwdth x = b p whch maxmzes ts ndvdual surplus accordng to equaton 4. The dscusson so far assumes that each prce p s per unt average bandwdth. A prce based on unt equvalent bandwdth [3] may be farer snce t takes nto account the burstness of user traffc. In ths case, the user adaptaton of the source rate s more complcated. If effectve bandwdth s used, a user could calculate a new average bandwdth when the prce ncreases. Alternatvely, t could ntroduce addtonal bufferng at the source to reduce ts burstness, at the cost of a hgher delay, thus reducng the effectve bandwdth. III. PRICING STRATEGIES A few prcng schemes are wdely used n the Internet today [4]: access-rate-dependent charge (AC), volume-dependent charge (V), or the combnaton of the both (AC-V). An AC chargng scheme s usually one of two types: allowng unlmted use, or allowng lmted duraton of connecton, and chargng a per-hour fee for addtonal connecton tme. Smlarly, AC-V chargng schemes normally allow some amount of volume to be transmtted for a fxed access fee, and then mpose a per-volume charge. Although tme-of-day dependent chargng s commonly used n telephone networks, t s not generally used n the current Internet. User experments [5] ndcate that usage-based prcng s a far way to charge people and allocate network resources. Both connecton tme and the transmtted volume reflect the usage of the network. Chargng based on connect-tme only works when resource demands per tme unt are roughly unform. Snce ths s not the case for

4 IEEE INFOCOM 4 Internet applcatons and across the range of access speeds, we only consder volume-based chargng. In ths paper, we study two knds of volume-based prcng: a fxed-prce (FP) polcy wth a fxed unt volume prce, and a congeston-prce-based adaptve servce (CPA) n whch the unt volume prce has a congeston-senstve component. In the fxed prce model, the network charges the user per volume of data transmtted, ndependent of the congeston state of the network. The per-byte charge can be the same for all servce classes ( flat, FP-FL), depend on the servce class (FP-PR), depend on the tme of day (FP-T) or a combnaton of tmeof-day and servce class (FP-PR-T). If the prce does not depend on the congeston condtons n the network, customers wth less bandwdth-senstve applcatons have no motvaton to reduce ther traffc as network congeston ncreases. As a result, ether the servce request blockng rate wll ncrease at the call admsson control level, or the packet delay and droppng rate wll ncrease at the queue management level. Havng a congeston-dependent component n the servce prce provdes a monetary ncentve for adaptve applcatons to adapt ther servce class and/or sendng rates accordng to network condtons. In perods of resource scarcty, qualty senstve applcatons can mantan ther resource levels by payng more, and relatvely qualtynsenstve applcatons wll reduce ther sendng rates or change to a lower class of servce. The total prce conssts of a congeston-dependent component and a fxed volumebased charge. The fxed volume-based charge has the same 4 chargng modes as n FP, gvng the prcng models CP-FL, CP-PR, CP-T, CP-PR-T. A. Proposed Prcng Scheme We assume that routers support multple servce classes and that each router s parttoned to provde a separate lnk bandwdth and buffer space for each servce, at each port. We use the framework of the compettve market model [6]. The compettve market model defnes two knds of agents: consumers and producers. Consumers seek resources from producers, and producers create or own the resources. The exchange rate of a resource s called ts prce. The routers are consdered the producers and own the lnk bandwdth and buffer space for each output port. The flows (ndvdual flows or aggregate of flows) are consdered consumers who consume resources. The congeston-dependent component of the servce prce s computed perodcally, wth a prce computaton nterval τ. The total demand for lnk bandwdth s based on the aggregate bandwdth reserved on the lnk for a prce computaton nterval, and the total demand for the buffer space at an output port s the average buffer occupancy durng the nterval. The supply bandwdth and buffer space need not be equal to the nstalled capacty; nstead, they are the targeted bandwdth and buffer space utlzaton. The congeston prce wll be leved once demands exceeds a provder-set fracton of the avalable bandwdth or buffer space. We now dscuss the formulaton of the fxed charge, whch we decompose nto holdng charge and usage charge, and the formulaton of the congeston charge. ) Holdng Charge: A servce may enforce admsson control to ensure some level of performance. In ths case, the applcatons admtted nto the network wll mpose some potental cost by deprvng other applcatons the opportunty to be admtted. Hence, t s far to charge the admtted applcatons a holdng prce. The holdng charge can be calculated based on the followng consderaton. If a partcular flow or flow-aggregate does not utlze the resources (buffer space or bandwdth) set asde for t, we assume that the scheduler allows the resources to be used by excess traffc from a lower level of servce. The holdng charge reflects revenue lost by the provder because nstead of sellng the allotted resources at the usage charge of the gven servce level (f all of the reserved resources were consumed) t sells the reserved resources at the usage charge of a lower servce level. The holdng prce (p j h ) of a servce class j s therefore set to be proportonal to the dfference between the usage prce for that class and the usage prce for the next lower servce class. The holdng prce can be represented as: p j h = αj (p j u p j u ), (8) where α j s a scalng factor related to servce class j. The holdng charge c j h (n) when the customer reserves a bandwdth r j (n) from class j s gven by: c j h (n) =pj h rj (n)τ j (9) where τ j s the negotaton perod for class j. r j (n) can be a bandwdth requrement specfed explctly by the customer, or estmated from the traffc specfcaton and servce request of the customer. ) Usage Charge: The usage charge s determned by the actual resources consumed, the average user demand, the level of servce guaranteed to the user, and the elastcty of the traffc. The usage prce (p u ) wll be set such that t allows a retal network to recover the cost of the purchase from the wholesale market, and varous fxed costs assocated wth the servce. In a network supportng multple classes of servce, the dfference n the charge between dfferent servce classes would presumably depend on the dfference n performance between the classes. The model we consder s a network supportng J classes of servces, the servce prce for class j s p j u, the long tme user bandwdth demand s known (e.g., through statstcs) and can be represented as x j (p u,p u,..., p J u), and the cost of havng capacty C durng one unt of tme s f(c). The provder s decson problem s to choose the optmal prces for each class that optmze ts proft: J max[ x j (p u,p u,..., p J u)p j u f(c)], p j u j subject to: r(x j (p u,p u,..., p J u)) R, j J () where r represents the bandwdth requrement for all classes, and R s the total bandwdth avalablty of the network. Assumng users choose servce classes ndependently, the total demand for a class over a long enough tme perod depends only on the prce for that class. If we assume the users have the utlty functons of Secton II, the total demand of servce class j can be represented as a constant elastcty model:

5 IEEE INFOCOM 5 x j (p j u ) = Aj /p j u, whch vares nversely wth the prce of the servce class. A j reflects the total wllngness to pay of users belongng to servce class j. Servce prcng for dfferentated servce DffServ supports SLA negotaton between the user and the network. An SLA generally ncludes traffc parameters, whch descrbe the user s traffc profle, and performance parameters, whch characterze the level of performance that the network promses to provde to the conformng part of the user s traffc. A wdely used descrptor for a user s traffc profle conssts of a peak rate, a sustanable rate, and a maxmum burst tolerance. The generally consdered QoS parameters are delay and loss. Mechansms, such as weghted far queung (WFQ) and class based queung (CBQ) can be used to provson resources for dfferent servce classes. In general, a class wth lower load leads to lower delay expectaton. A hgher level of servce class s expected to have a lower average load, and hence lower average delay. If we do not consder the dfference n element costs for dfferent classes, chargng servces proportonal to ther ndvdual expected load seems to reasonably reflect the cost of provdng the servces and the dfferences between ther performance. Assumng that unt bandwdth of a servce class would be charged a basc rate p basc f all ts bandwdth were used, and the expected load rato of servce class j s ρ j,the unt bandwdth prce for servce class j can then be estmated as p j u = p basc/ρ j. The effectve bandwdth consumpton of an applcaton wth rate x j can be represented as x j /ρ j. For constant elastcty demand, x j (p j u ) = Aj /p j u, and the effectve bandwdth consumpton s A j /(p j u ρj ). Then the prce optmzaton problem of equaton can be wrtten as: max[ p j u J j A j p j u p j u f(c)], subject to: p j u = p basc ρ j, J j A j p j uρ j C () The Lagrangan for the problem can be represented as: J J max [ A j j + λ(c Aj ) f(c)] () p basc p basc j The optmal soluton s: J j p basc = Aj C, pj u = p J basc j = Aj (3) ρ j Cρ j The bandwdth provsoned for each servce class wll be gven by A j /p basc, and s hence proportonal to total user wllngness to pay for that class. The usage charge c j u (n) for class j over a perod n n whch v j (n) bytes were transmtted s gven by: c j u (n) =p j uv j (n) (4) 3) Congeston Charge: A smple usage-based chargng scheme montors the data volume transmtted and n prncple charges users based on ther average rate. Chargng accordng to the mean rate, though encouragng the user to use network bandwdth more effcently, does not dscourage users from selectng large traffc contracts and sendng the worst-case traffc allowed by ther contract, whch create problems for network traffc management. An approprate prcng scheme should provde users the ncentves to select traffc contracts that reflect ther actual needs. Effectve bandwdth [9][7] and prcng based on effectve bandwdth [3] have been proposed n a multple-servce-class envronment. However, effectve bandwdth normally accounts for the worst case traffc subject to the traffc profle of the SLA. The contract for typcal users has an effectve bandwdth much larger than the mean rate. Provsonng based on equvalent bandwdth s not economcally effcent n a DffServ envronment. Performance guarantees n DffServ are qualtatve and can be very loose. Ths may make t dffcult to evaluate the equvalent bandwdth. Also, DffServ does not allocate resources to applcatons based on ther effectve bandwdth. Therefore, t appears unfar to charge users based on ther profle declaraton only, though the charge should take the profle nto account. To encourage users to reduce ther resource requrements under network resource contenton, we propose an addtonal congestonsenstve prce component under these condtons. The general network resources consdered are bandwdth and buffer space. Two knds of congeston prcng can be consdered: prcng when the expected load bound s exceeded, or prcng when buffer occupancy reaches certan level. In the frst case, when the average demand for a certan class exceeds a threshold, an addtonal congeston prce s charged all users of that class. In the case of prorty droppng for AF class, the droppng precedence s only consdered when the buffer occupancy reaches dfferent thresholds. The same thresholds can be assocated wth dfferent congeston or buffer prces. When each threshold s reached, user packets wth the correspondng precedence level begn to be dropped wth a certan probablty, and users wth hgher precedence levels are charged the addtonal buffer prce. Therefore, the hgher precedence users pay the sum of buffer prces correspondng to all the exceeded thresholds. Durng congeston, lower precedence users wll suffer lost packets, or reduce ther rate, or smoothen ther traffc at the source (at the cost of hgher delay due to bufferng), or change to a hgher precedence and pay a hgher prce. Both knds of congeston prce for a servce class can be calculated as an teratve tâtonnement process [6]: p j c(n) =mn[{p j c(n ) + σ j (D j,s j )(D j S j )/S j, } +,p j max] (5) where D j and S j represent the current total demand and supply respectvely, and σ j s a factor used to adjust the convergence rate. σ j maybeafunctonofd j and S j ;nthat case, t would be hgher when congeston s severe. D j and S j wll be dfferent for bandwdth and buffer space congeston. The router begns to apply the congeston charge only when the total demand exceeds the supply. Even after the congeston s removed, a non-zero, but gradually decreasng congeston charge s appled untl t falls to zero to protect aganst further congeston. In our smulatons, we also used a prce adjustment threshold parameter θ j to lmt the frequency wth whch the prce s updated. The congeston prce s updated f the calculated prce ncrement exceeds θ j p j c (n ). The maxmum

6 IEEE INFOCOM 6 congeston prce s bounded by the p j max. When a servce class needs admsson control, all new arrvals are rejected when the prce reaches p max j.ifp j c reaches p j max frequently, t ndcates that more resources are needed for the correspondng servce, or usage prce for a class needs to be adjusted to reflect the new demand statstcs. For a perod n, the total congeston charge s gven by c j c (n) =p j c(n)v j (n). (6) Based on the prce formulaton strategy descrbed above, a router arrves at a cost structure for a partcular RNAP flow or flow-aggregate at the end of each prce update nterval. The total charge for a sesson s gven by c j s = N [p j h rj (n)τ j +(p j u + p j c(n))v j (n)] (7) n= where N s the total number of ntervals spanned by a sesson. In some cases, the network may set the usage charge to zero, mposng a holdng charge for reservng resources only, and/or a congeston charge durng resource contenton. Also, the holdng charge would be set to zero for servces wthout explct resource reservaton or admsson control, for example, best effort servce. Snce the re-negotaton of network servces wll generally be drven by prce changes, the stablty of the negotaton process s dscussed n related work wth a greater focus on prcng [8]. IV. RESOURCE NEGOTIATION THROUGH RNAP The prcng algorthms and adaptaton framework presented n ths paper do not depend on any partcular network archtecture or protocol. However n ths paper, we smulated our results n an envronment supportng dynamc servce negotaton through the Resource Negotaton and Prcng protocol (RNAP) [5][8], usng a centralzed (RNAP-C) network management archtecture. We frst brefly revew the RNAP framework, and then descrbe the prcng and charge formulaton process used. In the RNAP framework, we assume that the network makes servces wth certan QoS characterstcs avalable to user applcatons, and charges prces for these servces that, n general, vary wth the avalablty of network resources. Network resources are obtaned by user applcatons through negotaton between the Host Resource Negotator (HRN) on the user sde, and a Network Resource Negotator (NRN) actng on behalf of the network. The HRN negotates on behalf of one or multple applcatons belongng to a multmeda system. In an RNAP sesson, the NRN perodcally provdes the HRN updated prces for a set of servces. Based on ths nformaton and current applcaton requrements, the HRN determnes the current optmal transmsson bandwdth and servce parameters for each applcaton. It re-negotates the contracted servces by sendng a Reserve message to the NRN, and recevng a Commt message as confrmaton or denal. The HRN only nteracts wth the local NRN. If ts applcaton flows traverse multple domans, resource negotatons are extended from end to end by passng RNAP messages hop-by-hop from the frst-hop NRN untl the destnaton COPS messages Table Table Doman Routng Table Resource Table B R R R R B Dest Next Hop Next Hop Next Hop (C, BW, Q, P) (C, BW, Q, P) (C, BW, Q, P) B R R B B, 3, 3, B3 R,, 3, B4 R, 3, 3, Step: determne a path (Table ) Step: accumulate prce along the path (Table ) Step 3: send total prce ($4/Mb) Fg.. B B3 R NRN B4 Prce formulaton n RNAP-C R B C: Servce class BW: average bandwdth (Mb) Q:average queue length P: prce ($/Mb) network NRN, and vce versa. End-to-end prces and charges are computed by accumulatng local prces and charges as Quotaton and Commt messages travel hop-by-hop upstream towards the HRN. The NRN mantans local state nformaton for a doman for chargng and other purposes. It makes the admsson decson and decdes the prce for a servce, based on the servce specfcatons alone, or by also takng nto account routng and confguraton polces, and network load. In the latter case, the NRN sts at a router that belongs to a lnk-state routng doman (for example an OSPF area) and has an dentcal lnk state database as other routers n the doman. Ths allows t to calculate all the routng tables of all other routers n the doman usng Djkstra s algorthm. The NRN mantans a doman routng table whch fnds any flow route that ether ends n ts own doman, or uses ts doman as a transt doman (Fg. ). The doman routng table wll be updated whenever the lnk state database s changed. A NRN also mantans a resource table, whch allows t to keep track of the avalablty and dynamc usage of the resources (bandwdth, buffer space). In general, the resource table stores resource nformaton for each servce provded at a router. The resource table allows the NRN to compute a local prce at each router (for nstance, usng the usage-based prcng strategy descrbed n Secton III). For a partcular servce request, the NRN frst looks up the path on whch resources are requested usng the doman routng table, and then uses the per-router prces to compute the accumulated prce along ths path. The resource table also facltates montorng and provsonng of resources at the routers. To enable the NRN to collect resource nformaton, routers n the doman perodcally report local state nformaton (for nstance, average buffer occupancy and bandwdth utlzaton) to the NRN. In ths paper, we extend COPS [9] for ths purpose. To compute the charge for a flow, ngress routers mantan per-flow (or aggregated flow from neghborng doman) state nformaton about the data volume transmtted durng a negotaton perod. Ths nformaton s perodcally transmtted to the NRN, allowng the NRN to compute the charge for the perod. The NRN uses the computed prce and charge to mantan chargng state nformaton for each RNAP sesson. A network doman manages ts own prcng scheme (whch

7 IEEE INFOCOM 7 may be congeston senstve or statc) ndependent of other domans, and wll have ts own per unt resource costs for each class. When an user flow traverses multple domans, RNAP messagng collates prcng and bllng nformaton from each doman and determne the total prce/charge for the user. For reducng the overhead due to per-flow RNAP message processng and storage, we consder a snk-tree based aggregaton scheme n [8]. The RNAP messages and state nformaton are aggregated n the core networks, allowng data measurement and chargng to be at much larger granularty. V. SIMULATION MODEL In ths secton, we descrbe our smulaton model for the CPA and FP polces. We smulate a sngle DffServ servce doman, under whch resources are not explctly reserved for each flow. We smulate the servce performance wth or wthout admsson control from the doman. User resource requrements are declared explctly through RNAP, allowng admsson control to be enforced f requred n an experment. The ndvdual and total user resource demands are also obtaned through measurement. Prce and network statstcs are sgnaled to users through RNAP. We used the network smulator [] envronment to smulate two network topologes, shown n Fg. and Fg. 3. Topology contans two backbone nodes, sx access nodes, and twentyfour end nodes. Topology two contans fve backbone nodes, ffteen access nodes, and sxty end nodes. Topology two was also used n []. All lnks are full duplex and pont-to-pont. The lnks connectng the backbone nodes are 3 Mb/s, the lnks connectng the access nodes to the backbone nodes are Mb/s, and the lnks connectng the end nodes to the access nodes are Mb/s. At each end node, there s a fxed number N s of sendng users. We use topology n most of our smulatons to allow congeston to be smulated at a sngle bottleneck node, and use topology to llustrate the CPA performance under a more general network topology [8]. We modfed the DffServ module developed by Sean Murphy to support dynamc SLA negotaton, and montor the user traffc at ngress pont. A Weghted-Round-Robn scheduler s modeled at each node, wth weghts dstrbuted equally among EF, AF, and Best Effort (BE) classes. Although the DffServ proposals menton 4 AF classes wth three levels of drop precedence n each, we only smulated one AF class to make the smulatons less resource-ntensve, snce ths does not affect the general results n any way. Three dfferent buffer management algorthms are used for dfferent DffServ classes - tal-droppng for EF, RED-wth-In-Out [] for AF, and Random Early Detecton [3] for the BE traffc. The default queue length for EF, AF and BE are set respectvely to 5,, packets. Other parameters are set to the default values n the network smulator mplementaton. A combnaton of exponental on-off and Pareto on-off traffc sources are used n the smulaton. Unless otherwse specfed, the traffc conssts of 5% of each for all the servce classes, and the on tme and off tme are both set to.5 seconds. The shape parameter for Pareto sources s set to.5. The mean packet sze s set to bytes. The traffc Senders Mb/s A A A3 B Mb/s 3 Mb/s Fg.. Smulaton network topology Mb/s A A A3 Mb/s B B B A5 A4 A3 3 Mb/s B5 A4 A5 A6 A7 A8 A9 Fg. 3. Smulaton network topology B4 B3 A6 A5 A4 Recevers condtoners are confgured wth one profle for each traffc source, wth peak rate and bucket sze set to the n-off source peak rate and maxmum amount of traffc sent durng an on perod respectvely for both EF and AF classes. We also characterze the system load by burst ndex and offered load. The burst ndex s defned as OffTme/(OnTme +OffTme)for both types of On-Off sources. The offered load for a servce class s defned as the rato between the total user resource requrement for a servce type, and the confgured class capacty at the bottleneck. Under the FP polcy, the total user resource requrement s also the actual resource demand from all the users. Under the CPA polcy, the total user resource requrement s what the total resource demand would be f there were no resource contenton at the bottleneck and the network dd not mpose an addtonal congestondependent prce. User requests are generated accordng to a Posson arrval process and the lfetme of each flow s exponentally dstrbuted wth an average length of mnutes. In topology, users from the sender sde ndependently ntalze undrectonal flows towards randomly selected recever sde end nodes. N s flows wll be ntalzed at one node. At most N s flows (6 sessons wth N s set to 5) can run smultaneously n the whole network. In topology, all the users ntalze undrectonal flows towards randomly selected end nodes. At most 6N s users (36 sessons wth N s set to 6) are allowed to run smultaneously n the whole network. For ease of understandng, all prces n ths secton are gven n terms of prce per mnute of a 64 kb/s transmsson, currently equvalent to a telephone call. The basc prce charged by the FP polcy, and the basc usage prce charged by CPA (p basc ), are both set to $.8/mn. We set the target average load of the EF class at 4%, the AF class at 6%, and the BE class at 9%. Therefore, based on the prcng strategy proposed n Secton III, the usage prce for EF, AF and BE classes are set respectvely as $./mn, $.3/mn, and $.89/mn. When admsson control s enforced, the holdng A A A

8 IEEE INFOCOM 8 prce for the CPA polcy s correspondngly set to $.67/mn for EF class, and $.44/mn for AF class. Congeston prcng s appled when nstantaneous usage exceeds the target load threshold of each class or when the loss or delay exceeds /3 of the bounds at a node assocated wth the class (delay bound of ms, 5 ms, and ms respectvely for EF, AF, and BE, and loss bounds of 6, 4 and respectvely). The prce adjustment procedure s also controlled by a par of parameters, the prce adjustment step σ from equaton 5 and the prce adjustment threshold parameter θ, defned n Secton III. Unless otherwse specfed, values of σ =.6 and θ =.5 are used. The users are assumed to have the general form of the utlty functon shown n Secton II. At the begnnng of each experment, the user populaton s dvded nto users of the EF, AF and BE classes, although n some experments they are allowed to adapt to prce changes by swtchng to a dfferent class. For EF users, the elastcty factor factor w (whch s also the user s wllngness to pay), s unformly dstrbuted between $.3/mn and $.4/mn for a 64 kb/s bandwdth. For AF and BE users, t s unformly dstrbuted between $.9/mn and $.6/mn, and $.6/mn and $.8/mn respectvely. The mnmum delay and loss requrements for each type of users are set to be the same as the expected performance bound of the correspondng servce class. The opportunty cost parameter U s set to the amount a user s wllng to pay for ts mnmum bandwdth requrement, and s hence gven by U = p hgh x mn,wherep hgh s the maxmum prce the user wll pay before termnatng hs connecton altogether. Users re-negotate ther resource requrements wth a perod of 3 seconds n all the experments. The total smulaton tme for each experment s, seconds. We use a number of engneerng and economc metrcs to evaluate our experments. The engneerng metrcs nclude the average traffc arrval rate at the bottleneck, the average packet delay, the average packet loss rate, and the user request blockng probablty. The averages are computed as exponentally weghted movng averages. The economc performance metrcs nclude the average user beneft (the perceved value obtaned by users based on ther utlty functons), the end-toend prce for each servce class. VI. RESULTS AND DISCUSSION In ths secton, we smulate the FP polcy and CPA polcy under dentcal traffc condtons, and compare the relatve performance. For ease of presentaton, a sngle traffc parameter for the AF class was vared n each experment, and ts effect on CPA and FP polcy performance was studed. We conducted four groups of experments. In the frst and second groups, we vary the load burstness and average load respectvely of the AF class, and evaluate the mprovements gven by CPA over FP. In the thrd experment, ncentve drven traffc mgraton between classes s shown to mprove the overall system performance. In the last experment, we show that access control to a servce class s crtcal n mantanng expected performance (a) (c) (e) Bottleneck traffc arrval rate Average packet delay (second) average prce prce devaton Burst ndex x AF burst ndex AF burst ndex (b) (d) (f) Average packet loss rate Average user beneft ($/mn) burst ndex:.4 burst ndex:.6 burst ndex: tme (s) x 4 x AF burst ndex AF burst ndex Fg. 4. System dynamcs under CPA wth ncrease n AF traffc burst ndex: (a) prce average and standard devaton of AF class; (b) varaton over tme of AF. Performance metrcs of CPA and FP polces as a functon of burst ndex of AF class: (c) average packet delay; (d) average packet loss; (e) average traffc arrval rate; (f) average user beneft. levels. Combnng access control wth user servce adaptaton effectvely reduces the request blockng rate. A. Effect of Traffc Burstness We frst compare the performance of FP and CPA polces as the burst ndex of AF class ncreases, at a constant average offered load of 6%. Fg. 4 (a) shows that the average AF prce ncreases under CPA due to the ncreasng congeston prce as the burst ndex exceeds.4. In response, the AF traffc backs off. Fg. 4 (a) also shows that the standard devaton n the AF prce ncreases wth the burst ndex, ndcatng greater fluctuatons n the prce. Fg. 4 (b) shows the dynamc varaton of the AF class prce at three dfferent levels of burstness, confrmng ths trend. Fg. 4 (c) and (d) show that under FP polcy the average packet delay and loss of the AF class ncrease sharply as the burst ndex exceeds.4. As a result of the user traffc back-off under CPA the delay and loss of AF class are well controlled below the respectve performance bounds of 5 ms and 4 up to a burst ndex of.8. The average user beneft for CPA (Fg. 4 f) decreases due to the reducton of bandwdth, but remans hgher than that of the FP polcy. There s also a smaller degradaton n the performance of the BE class at hgh burst ndces. Ths appears to be because the BE class operates under a relatvely hgh load, and therefore borrows bandwdth from the AF class when the AF class s lghtly loaded. It can no longer do so when the AF traffc burstness ncreases.

9 IEEE INFOCOM 9 (a) average prce prce devaton (b) offered load:.8 offered load:. offered load: tme (s) x 4 (a) offered load:.8 offered load:. offered load: tme (s) x 4 (b) Rato of AF traffc mgratng between classes EF AF BE Average packet delay (second).5..5 Average packet loss Average packet delay (second) 9 x Average packet loss rate x (c) (e) Bottleneck traffc arrval rate (d) (f) Average user beneft ($/mn) Fg. 5. System dynamcs under CPA wth ncrease n : (a) average and standard devaton of AF class prce; (b) varaton over tme of AF class prce. Performance metrcs of CPA and FP polces as a functon of : (c) average packet delay; (d) average packet loss; (e) average bottleneck traffc arrval rate; (f) average user beneft. The results n ths secton ndcate that the CPA polcy takes advantage of applcaton adaptvty for sgnfcant gans n network performance, and perceved user beneft, relatve to the fxed-prce polcy. The congeston-based prcng s stable and effectve. B. Effect of Traffc Load In ths smulaton, we keep the load and burstness of EF class and BE class and the burst ndex of the AF class at ther default values, and vary the offered load of AF class. The average AF prce under CPA s seen to ncrease wth offered load (Fg. 5 (a)). The standard devaton of the prce shows an ncrease to a certan level and then a decrease. Intally, the prce devaton ncreases due to the more aggressve congeston control. At heavy loads, the ncreased multplexng of user demand smooths the total demand, and therefore reduces fluctuatons n the prce. Fg. 5 (e) shows that the actual arrval rate of AF under CPA backs off as users adapt to the hgher prce. Fgs. 5 (c) and (d) show that the delay and loss of AF class under FP quckly ncreases after the offered load ncreases above.6 and approaches the provsoned capacty. As a result, the performance bounds for AF class can no longer be met. The hgh AF load also degrades BE performance. Ths s apparently because BE operates at a hgh load (.9) and tends to borrow bandwdth from AF and EF when the latter classes are lghtly loaded. Fgs. 5 (c), (d), and (e) show that CPA coupled wth user (c) (d) Fg. 6. Performance metrcs of CPA and FP polces wth traffc mgraton between classes: (a) varaton over tme of AF class prce; (b) rato of AF class traffc mgratng through class re-selecton; (c) average packet delay of all classes; (d) average packet loss of all classes; adaptaton s able to control congeston and mantan the total traffc load of a servce class at the targeted level, and hence allows the servce class to meet the expected performance bounds. Smlar to our observaton n Secton VI-A, f the nomnal prce of the system correctly reflect long-term user demand, dynamc prcng drven servce re-negotaton can effectvely lmts short-term fluctuatons n load. Usage prce of a class should be adjusted f persstent hgh user demand exst for a servce. C. Load Balance between Classes As seen from the prevous secton, the performance of a class wll suffer f the load nto that class s too hgh. In general, a user under CPA polcy wll select a servce class whch provdes t the hghest beneft based on the prce and performance parameters of a class as announced by the provders. The performance parameters are generally based on long-term statstcs. In ths secton, we assume that a user can learn from network performance data receved over a short perod, and select the class that would provde the hghest beneft based on the user utlty functon, network performance statstcs and servce prce, as dscussed n Secton II. In ths smulaton, the EF and BE classes are loaded at 3% and 8% respectvely. When the load of AF class ncreases, the performance of AF class degrades and congeston prce s nvoked. In response, some applcatons swtch from the AF class to the EF class, whch provdes better performance guarantee, or BE class, whch allows t more bandwdth at a cheaper prce. As the result of ths re-selecton, the load s better balanced across classes, and overall performance of the system mproves (Fg. 6 (c) and (d)). Fg. 6 (a) shows that wth load balancng n combnaton wth adaptaton wthn a sngle class, the congeston prce needs to be nvoked much less often than wth adaptaton wthn a class only, as n Fg. 5 (b). The proporton of mgratng traffc s shown n Fg. 6 (b). We see even when a small porton of users select other

10 IEEE INFOCOM (a) (c) Average packet delay (second) AF average prce prce devaton x (b) (d) Request blockng rate Average packet loss x Fg. 7. System dynamcs under CPA wth access control CPA as AF offered load ncreases: (a) average and standard devaton of AF class prce. Performance metrcs of CPA and FP polces wth access control as a functon of : (b) user requests blockng rate; (c) average packet delay; (d) average packet loss. servce classes, the performance of the over-loaded class s greatly mproved. D. Effect of Admsson Control We have seen that the performance of a class can not be expected wthout any access control. In ths secton, we compare the performance of FP and CPA for a network wth admsson control for EF and AF class. The admsson threshold for each class s set to.5 tmes the target load to ncrease the effcency of the network. Wth admsson control, the performance of EF and AF classes are well controlled (Fg. 7 c and d). However, due to the burstness of the traffc, the blockng rate under FP s hgh even at a very small offered load (Fg. 7 b), and ncreases almost lnearly as the offered load ncreases beyond.6. Wth congeston control and servce contract re-negotaton, the blockng rate of CPA s seen to be up to 3 tmes smaller than that under the FP polcy, and actually starts to decrease after reachng a maxmum at offered load.8. Ths s because the prce adjustment step s proportonal to the excess bandwdth above the targeted utlzaton and ncreases progressvely faster wth offered load at hgher loads, and the user bandwdth request decreases proportonally wth the prce accordng to the general utlty functon of Secton II. Compared to Secton VI-B, the average prce under CPA (Fg. 7 a) s bounded to a smaller value at hgh offered loads, and has a smaller fluctuaton. The results ndcate that access control s mportant n mantanng the expected performance of a class. However, admsson control by tself may lead to a hgh blockng rate due to the network dynamcs. By combnng admsson control wth user traffc adaptaton, the network s more effcently used. Wth admsson control, the dynamcs of the network prce can also be better controlled, so that users have a more relable expectaton of the prce. VII. RELATED WORK Mcroeconomc prncples have been appled to varous network traffc management problems. The studes n [4][5][6] are based on a maxmzaton process to determne the optmal resource allocaton such that the utlty (a functon that maps a resource amount to a satsfacton level) of a group of users s maxmzed. In [7][8][6][9], the resources are prced to reflect demand and supply. Some of these methods are lmted by ther relance on a well-defned statstcal model of source traffc, and are generally not ntended to adapt to changng traffc demands. The study n [7] shows that compare to tradtonal flat prcng, servce-class senstve prcng results n hgher network performance. Prcng for DffServ has also been studed n [3] through equvalent bandwdth. As has been ponted out earler, equvalent bandwdth may be too conservatve for resource provsonng n a DffServ envronment, and hence prcng based on equvalent bandwdth may not be far to the users. Also, t s not trval for users to adapt ther requrements dynamcally to meet ther equvalent bandwdth constrants. Although there s some overlap between the cted work and ours, our work s drected to studyng and solvng somewhat dfferent problems - developng a prcng model for DffServ, and studyng DffServ performance n a dynamc servce and prce negotaton envronment. VIII. SUMMARY In ths work, we have developed a reasonably complete DffServ prcng model. We have proposed a prce structure for dfferent servce classes n DffServ based on ther relatve performance, long-term demand, and short-term fluctuatons n demand. We have ntegrated ths prcng model nto a dynamc servce negotaton envronment n whch servce prces ncrease n response to congeston, and users adapt to prce ncreases by adaptng ther sendng rate and/or choce of servce. We have also modeled the demand behavor of adaptve users based on a physcally reasonable user utlty functon. Our smulaton results show that the dfferent DffServ classes provde dfferent levels of servce only when they operate at dfferent target utlzaton. In the absence of explct admsson control, a servce class loaded beyond ts target utlzaton (under ether sustaned or bursty loads) no longer meets ts expected performance levels. Under these condtons, a congeston-senstve prcng polcy (CPA) coupled wth user rate adaptaton s able to control congeston and allow a servce class to meet ts performance assurances under large or bursty offered loads. Users see a reasonably stable servce prce and are able to mantan a very stable expendture. Allowng users to mgrate between servce classes n response to prce ncrease and network performance further stablzes the ndvdual servce prces whle mantanng the system performance. When admsson control s enforced beyond a threshold load for each class, performance bounds can be met wth a fxed servce prce. However, n ths case, the CPA polcy

Simulation Based Analysis of FAST TCP using OMNET++

Simulation Based Analysis of FAST TCP using OMNET++ Smulaton Based Analyss of FAST TCP usng OMNET++ Umar ul Hassan 04030038@lums.edu.pk Md Term Report CS678 Topcs n Internet Research Sprng, 2006 Introducton Internet traffc s doublng roughly every 3 months

More information

Efficient Distributed File System (EDFS)

Efficient Distributed File System (EDFS) Effcent Dstrbuted Fle System (EDFS) (Sem-Centralzed) Debessay(Debsh) Fesehaye, Rahul Malk & Klara Naherstedt Unversty of Illnos-Urbana Champagn Contents Problem Statement, Related Work, EDFS Desgn Rate

More information

Internet Traffic Managers

Internet Traffic Managers Internet Traffc Managers Ibrahm Matta matta@cs.bu.edu www.cs.bu.edu/faculty/matta Computer Scence Department Boston Unversty Boston, MA 225 Jont work wth members of the WING group: Azer Bestavros, John

More information

Wishing you all a Total Quality New Year!

Wishing you all a Total Quality New Year! Total Qualty Management and Sx Sgma Post Graduate Program 214-15 Sesson 4 Vnay Kumar Kalakband Assstant Professor Operatons & Systems Area 1 Wshng you all a Total Qualty New Year! Hope you acheve Sx sgma

More information

Scheduling and queue management. DigiComm II

Scheduling and queue management. DigiComm II Schedulng and queue management Tradtonal queung behavour n routers Data transfer: datagrams: ndvdual packets no recognton of flows connectonless: no sgnallng Forwardng: based on per-datagram forwardng

More information

Goals and Approach Type of Resources Allocation Models Shared Non-shared Not in this Lecture In this Lecture

Goals and Approach Type of Resources Allocation Models Shared Non-shared Not in this Lecture In this Lecture Goals and Approach CS 194: Dstrbuted Systems Resource Allocaton Goal: acheve predcable performances Three steps: 1) Estmate applcaton s resource needs (not n ths lecture) 2) Admsson control 3) Resource

More information

Real-Time Guarantees. Traffic Characteristics. Flow Control

Real-Time Guarantees. Traffic Characteristics. Flow Control Real-Tme Guarantees Requrements on RT communcaton protocols: delay (response s) small jtter small throughput hgh error detecton at recever (and sender) small error detecton latency no thrashng under peak

More information

Avoiding congestion through dynamic load control

Avoiding congestion through dynamic load control Avodng congeston through dynamc load control Vasl Hnatyshn, Adarshpal S. Seth Department of Computer and Informaton Scences, Unversty of Delaware, Newark, DE 976 ABSTRACT The current best effort approach

More information

Video Proxy System for a Large-scale VOD System (DINA)

Video Proxy System for a Large-scale VOD System (DINA) Vdeo Proxy System for a Large-scale VOD System (DINA) KWUN-CHUNG CHAN #, KWOK-WAI CHEUNG *# #Department of Informaton Engneerng *Centre of Innovaton and Technology The Chnese Unversty of Hong Kong SHATIN,

More information

Dynamic Bandwidth Provisioning with Fairness and Revenue Considerations for Broadband Wireless Communication

Dynamic Bandwidth Provisioning with Fairness and Revenue Considerations for Broadband Wireless Communication Ths full text paper was peer revewed at the drecton of IEEE Communcatons Socety subject matter experts for publcaton n the ICC 008 proceedngs. Dynamc Bandwdth Provsonng wth Farness and Revenue Consderatons

More information

RAP. Speed/RAP/CODA. Real-time Systems. Modeling the sensor networks. Real-time Systems. Modeling the sensor networks. Real-time systems:

RAP. Speed/RAP/CODA. Real-time Systems. Modeling the sensor networks. Real-time Systems. Modeling the sensor networks. Real-time systems: Speed/RAP/CODA Presented by Octav Chpara Real-tme Systems Many wreless sensor network applcatons requre real-tme support Survellance and trackng Border patrol Fre fghtng Real-tme systems: Hard real-tme:

More information

Analysis of Collaborative Distributed Admission Control in x Networks

Analysis of Collaborative Distributed Admission Control in x Networks 1 Analyss of Collaboratve Dstrbuted Admsson Control n 82.11x Networks Thnh Nguyen, Member, IEEE, Ken Nguyen, Member, IEEE, Lnha He, Member, IEEE, Abstract Wth the recent surge of wreless home networks,

More information

Virtual Machine Migration based on Trust Measurement of Computer Node

Virtual Machine Migration based on Trust Measurement of Computer Node Appled Mechancs and Materals Onlne: 2014-04-04 ISSN: 1662-7482, Vols. 536-537, pp 678-682 do:10.4028/www.scentfc.net/amm.536-537.678 2014 Trans Tech Publcatons, Swtzerland Vrtual Machne Mgraton based on

More information

Bandwidth Allocation for Service Level Agreement Aware Ethernet Passive Optical Networks

Bandwidth Allocation for Service Level Agreement Aware Ethernet Passive Optical Networks Bandwdth Allocaton for Servce Level Agreement Aware Ethernet Passve Optcal Networks Dawd Nowak Performance Engneerng Laboratory School of Electronc Engneerng Dubln Cty Unversty Emal: nowakd@eengdcue Phlp

More information

Parallelism for Nested Loops with Non-uniform and Flow Dependences

Parallelism for Nested Loops with Non-uniform and Flow Dependences Parallelsm for Nested Loops wth Non-unform and Flow Dependences Sam-Jn Jeong Dept. of Informaton & Communcaton Engneerng, Cheonan Unversty, 5, Anseo-dong, Cheonan, Chungnam, 330-80, Korea. seong@cheonan.ac.kr

More information

A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS

A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS Proceedngs of the Wnter Smulaton Conference M E Kuhl, N M Steger, F B Armstrong, and J A Jones, eds A MOVING MESH APPROACH FOR SIMULATION BUDGET ALLOCATION ON CONTINUOUS DOMAINS Mark W Brantley Chun-Hung

More information

ARTICLE IN PRESS. Signal Processing: Image Communication

ARTICLE IN PRESS. Signal Processing: Image Communication Sgnal Processng: Image Communcaton 23 (2008) 754 768 Contents lsts avalable at ScenceDrect Sgnal Processng: Image Communcaton journal homepage: www.elsever.com/locate/mage Dstrbuted meda rate allocaton

More information

DESIGNING TRANSMISSION SCHEDULES FOR WIRELESS AD HOC NETWORKS TO MAXIMIZE NETWORK THROUGHPUT

DESIGNING TRANSMISSION SCHEDULES FOR WIRELESS AD HOC NETWORKS TO MAXIMIZE NETWORK THROUGHPUT DESIGNING TRANSMISSION SCHEDULES FOR WIRELESS AD HOC NETWORKS TO MAXIMIZE NETWORK THROUGHPUT Bran J. Wolf, Joseph L. Hammond, and Harlan B. Russell Dept. of Electrcal and Computer Engneerng, Clemson Unversty,

More information

A fair buffer allocation scheme

A fair buffer allocation scheme A far buffer allocaton scheme Juha Henanen and Kalev Klkk Telecom Fnland P.O. Box 228, SF-330 Tampere, Fnland E-mal: juha.henanen@tele.f Abstract An approprate servce for data traffc n ATM networks requres

More information

Load-Balanced Anycast Routing

Load-Balanced Anycast Routing Load-Balanced Anycast Routng Chng-Yu Ln, Jung-Hua Lo, and Sy-Yen Kuo Department of Electrcal Engneerng atonal Tawan Unversty, Tape, Tawan sykuo@cc.ee.ntu.edu.tw Abstract For fault-tolerance and load-balance

More information

Priority-Based Scheduling Algorithm for Downlink Traffics in IEEE Networks

Priority-Based Scheduling Algorithm for Downlink Traffics in IEEE Networks Prorty-Based Schedulng Algorthm for Downlnk Traffcs n IEEE 80.6 Networks Ja-Mng Lang, Jen-Jee Chen, You-Chun Wang, Yu-Chee Tseng, and Bao-Shuh P. Ln Department of Computer Scence Natonal Chao-Tung Unversty,

More information

A New Token Allocation Algorithm for TCP Traffic in Diffserv Network

A New Token Allocation Algorithm for TCP Traffic in Diffserv Network A New Token Allocaton Algorthm for TCP Traffc n Dffserv Network A New Token Allocaton Algorthm for TCP Traffc n Dffserv Network S. Sudha and N. Ammasagounden Natonal Insttute of Technology, Truchrappall,

More information

CS 268: Lecture 8 Router Support for Congestion Control

CS 268: Lecture 8 Router Support for Congestion Control CS 268: Lecture 8 Router Support for Congeston Control Ion Stoca Computer Scence Dvson Department of Electrcal Engneerng and Computer Scences Unversty of Calforna, Berkeley Berkeley, CA 9472-1776 Router

More information

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr)

Helsinki University Of Technology, Systems Analysis Laboratory Mat Independent research projects in applied mathematics (3 cr) Helsnk Unversty Of Technology, Systems Analyss Laboratory Mat-2.08 Independent research projects n appled mathematcs (3 cr) "! #$&% Antt Laukkanen 506 R ajlaukka@cc.hut.f 2 Introducton...3 2 Multattrbute

More information

High Performance DiffServ Mechanism for Routers and Switches: Packet Arrival Rate Based Queue Management for Class Based Scheduling

High Performance DiffServ Mechanism for Routers and Switches: Packet Arrival Rate Based Queue Management for Class Based Scheduling Hgh Performance DffServ Mechansm for Routers and Swtches: Packet Arrval Rate Based Queue Management for Class Based Schedulng Bartek Wydrowsk and Moshe Zukerman ARC Specal Research Centre for Ultra-Broadband

More information

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields

A mathematical programming approach to the analysis, design and scheduling of offshore oilfields 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 A mathematcal programmng approach to the analyss, desgn and

More information

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique

The Greedy Method. Outline and Reading. Change Money Problem. Greedy Algorithms. Applications of the Greedy Strategy. The Greedy Method Technique //00 :0 AM Outlne and Readng The Greedy Method The Greedy Method Technque (secton.) Fractonal Knapsack Problem (secton..) Task Schedulng (secton..) Mnmum Spannng Trees (secton.) Change Money Problem Greedy

More information

Some Advanced SPC Tools 1. Cumulative Sum Control (Cusum) Chart For the data shown in Table 9-1, the x chart can be generated.

Some Advanced SPC Tools 1. Cumulative Sum Control (Cusum) Chart For the data shown in Table 9-1, the x chart can be generated. Some Advanced SP Tools 1. umulatve Sum ontrol (usum) hart For the data shown n Table 9-1, the x chart can be generated. However, the shft taken place at sample #21 s not apparent. 92 For ths set samples,

More information

Real-time interactive applications

Real-time interactive applications Real-tme nteractve applcatons PC-2-PC phone PC-2-phone Dalpad Net2phone vdeoconference Webcams Now we look at a PC-2-PC Internet phone example n detal Internet phone over best-effort (1) Best effort packet

More information

Technical Report. i-game: An Implicit GTS Allocation Mechanism in IEEE for Time- Sensitive Wireless Sensor Networks

Technical Report. i-game: An Implicit GTS Allocation Mechanism in IEEE for Time- Sensitive Wireless Sensor Networks www.hurray.sep.pp.pt Techncal Report -GAME: An Implct GTS Allocaton Mechansm n IEEE 802.15.4 for Tme- Senstve Wreless Sensor etworks Ans Koubaa Máro Alves Eduardo Tovar TR-060706 Verson: 1.0 Date: Jul

More information

A Binarization Algorithm specialized on Document Images and Photos

A Binarization Algorithm specialized on Document Images and Photos A Bnarzaton Algorthm specalzed on Document mages and Photos Ergna Kavalleratou Dept. of nformaton and Communcaton Systems Engneerng Unversty of the Aegean kavalleratou@aegean.gr Abstract n ths paper, a

More information

The Codesign Challenge

The Codesign Challenge ECE 4530 Codesgn Challenge Fall 2007 Hardware/Software Codesgn The Codesgn Challenge Objectves In the codesgn challenge, your task s to accelerate a gven software reference mplementaton as fast as possble.

More information

Mathematics 256 a course in differential equations for engineering students

Mathematics 256 a course in differential equations for engineering students Mathematcs 56 a course n dfferental equatons for engneerng students Chapter 5. More effcent methods of numercal soluton Euler s method s qute neffcent. Because the error s essentally proportonal to the

More information

An Entropy-Based Approach to Integrated Information Needs Assessment

An Entropy-Based Approach to Integrated Information Needs Assessment Dstrbuton Statement A: Approved for publc release; dstrbuton s unlmted. An Entropy-Based Approach to ntegrated nformaton Needs Assessment June 8, 2004 Wllam J. Farrell Lockheed Martn Advanced Technology

More information

Efficient Load-Balanced IP Routing Scheme Based on Shortest Paths in Hose Model. Eiji Oki May 28, 2009 The University of Electro-Communications

Efficient Load-Balanced IP Routing Scheme Based on Shortest Paths in Hose Model. Eiji Oki May 28, 2009 The University of Electro-Communications Effcent Loa-Balance IP Routng Scheme Base on Shortest Paths n Hose Moel E Ok May 28, 2009 The Unversty of Electro-Communcatons Ok Lab. Semnar, May 28, 2009 1 Outlne Backgroun on IP routng IP routng strategy

More information

MobileGrid: Capacity-aware Topology Control in Mobile Ad Hoc Networks

MobileGrid: Capacity-aware Topology Control in Mobile Ad Hoc Networks MobleGrd: Capacty-aware Topology Control n Moble Ad Hoc Networks Jle Lu, Baochun L Department of Electrcal and Computer Engneerng Unversty of Toronto {jenne,bl}@eecg.toronto.edu Abstract Snce wreless moble

More information

X- Chart Using ANOM Approach

X- Chart Using ANOM Approach ISSN 1684-8403 Journal of Statstcs Volume 17, 010, pp. 3-3 Abstract X- Chart Usng ANOM Approach Gullapall Chakravarth 1 and Chaluvad Venkateswara Rao Control lmts for ndvdual measurements (X) chart are

More information

Network-Driven Layered Multicast with IPv6

Network-Driven Layered Multicast with IPv6 Network-Drven Layered Multcast wth IPv6 Ho-pong Sze and Soung C. Lew Department of Informaton Engneerng, The Chnese Unversty of Hong Kong, Shatn, N.T., Hong Kong {hpsze8, soung}@e.cuhk.edu.hk Abstract.

More information

Shared Running Buffer Based Proxy Caching of Streaming Sessions

Shared Running Buffer Based Proxy Caching of Streaming Sessions Shared Runnng Buffer Based Proxy Cachng of Streamng Sessons Songqng Chen, Bo Shen, Yong Yan, Sujoy Basu Moble and Meda Systems Laboratory HP Laboratores Palo Alto HPL-23-47 March th, 23* E-mal: sqchen@cs.wm.edu,

More information

Advanced Computer Networks

Advanced Computer Networks Char of Network Archtectures and Servces Department of Informatcs Techncal Unversty of Munch Note: Durng the attendance check a stcker contanng a unque QR code wll be put on ths exam. Ths QR code contans

More information

A Hybrid Genetic Algorithm for Routing Optimization in IP Networks Utilizing Bandwidth and Delay Metrics

A Hybrid Genetic Algorithm for Routing Optimization in IP Networks Utilizing Bandwidth and Delay Metrics A Hybrd Genetc Algorthm for Routng Optmzaton n IP Networks Utlzng Bandwdth and Delay Metrcs Anton Redl Insttute of Communcaton Networks, Munch Unversty of Technology, Arcsstr. 21, 80290 Munch, Germany

More information

Transit Networking in ATM/B-ISDN based on Service Category

Transit Networking in ATM/B-ISDN based on Service Category Transt Networkng n ATM/B-ISDN based on Servce Category Young-Tak Km Dept. of Informaton and Communcaton Engneerng, College of Engneerng, YeungNam Unv. E-mal : ytkm@ynucc.yeungnam.ac.kr ABSTRACT The ATM

More information

Enhanced Signaling Scheme with Admission Control in the Hybrid Optical Wireless (HOW) Networks

Enhanced Signaling Scheme with Admission Control in the Hybrid Optical Wireless (HOW) Networks Enhanced Sgnalng Scheme wth Admsson Control n the Hybrd Optcal Wreless (HOW) Networks Yng Yan, Hao Yu, Henrk Wessng, and Lars Dttmann Department of Photoncs Techncal Unversty of Denmark Lyngby, Denmark

More information

Optimizing Document Scoring for Query Retrieval

Optimizing Document Scoring for Query Retrieval Optmzng Document Scorng for Query Retreval Brent Ellwen baellwe@cs.stanford.edu Abstract The goal of ths project was to automate the process of tunng a document query engne. Specfcally, I used machne learnng

More information

FAST TCP: Motivation, Architecture, Algorithms, Performance

FAST TCP: Motivation, Architecture, Algorithms, Performance FAST TCP: Motvaton, Archtecture, Algorthms, Performance Cheng Jn Davd X. We Steven H. Low Engneerng & Appled Scence, Caltech http://netlab.caltech.edu Abstract We descrbe FAST TCP, a new TCP congeston

More information

Quantifying Responsiveness of TCP Aggregates by Using Direct Sequence Spread Spectrum CDMA and Its Application in Congestion Control

Quantifying Responsiveness of TCP Aggregates by Using Direct Sequence Spread Spectrum CDMA and Its Application in Congestion Control Quantfyng Responsveness of TCP Aggregates by Usng Drect Sequence Spread Spectrum CDMA and Its Applcaton n Congeston Control Mehd Kalantar Department of Electrcal and Computer Engneerng Unversty of Maryland,

More information

Statistical Admission Control Using Delay Distribution Measurements

Statistical Admission Control Using Delay Distribution Measurements Statstcal Admsson Control Usng Delay Dstrbuton Measurements KARTIK GOPALAN State Unversty of New York at Bnghamton LAN HUANG IBM Almaden Research Center GANG PENG and TZI-CKER CHIUEH Stony Brook Unversty

More information

WIRELESS communication technology has gained widespread

WIRELESS communication technology has gained widespread 616 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 4, NO. 6, NOVEMBER/DECEMBER 2005 Dstrbuted Far Schedulng n a Wreless LAN Ntn Vadya, Senor Member, IEEE, Anurag Dugar, Seema Gupta, and Paramvr Bahl, Senor

More information

A Fair MAC Algorithm with Dynamic Priority for e WLANs

A Fair MAC Algorithm with Dynamic Priority for e WLANs 29 Internatonal Conference on Communcaton Software and Networks A Far MAC Algorthm wth Dynamc Prorty for 82.e WLANs Rong He, Xumng Fang Provncal Key Lab of Informaton Codng & Transmsson, Southwest Jaotong

More information

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz

Compiler Design. Spring Register Allocation. Sample Exercises and Solutions. Prof. Pedro C. Diniz Compler Desgn Sprng 2014 Regster Allocaton Sample Exercses and Solutons Prof. Pedro C. Dnz USC / Informaton Scences Insttute 4676 Admralty Way, Sute 1001 Marna del Rey, Calforna 90292 pedro@s.edu Regster

More information

Use of Genetic Algorithms in Efficient Scheduling for Multi Service Classes

Use of Genetic Algorithms in Efficient Scheduling for Multi Service Classes Use of Genetc Algorthms n Effcent Schedulng for Mult Servce Classes Shyamale Thlakawardana and Rahm Tafazoll Centre for Communcatons Systems Research (CCSR), Unversty of Surrey, Guldford, GU27XH, UK Abstract

More information

Intelligent Traffic Conditioners for Assured Forwarding Based Differentiated Services Networks 1

Intelligent Traffic Conditioners for Assured Forwarding Based Differentiated Services Networks 1 Intellgent Traffc Condtoners for Assured Forwardng Based Dfferentated Servces Networks B. Nandy, N. Seddgh, P. Peda, J. Ethrdge Nortel Networks, Ottawa, Canada Emal:{bnandy, nseddgh, ppeda, jethrdg}@nortelnetworks.com

More information

A Semi-Distributed Load Balancing Architecture and Algorithm for Heterogeneous Wireless Networks

A Semi-Distributed Load Balancing Architecture and Algorithm for Heterogeneous Wireless Networks A Sem-Dstrbuted oad Balancng Archtecture and Algorthm for Heterogeneous reless Networks Md. Golam Rabul Ala Choong Seon Hong * Kyung Hee Unversty, Korea rob@networkng.khu.ac.kr, cshong@khu.ac.kr Abstract

More information

Resource and Virtual Function Status Monitoring in Network Function Virtualization Environment

Resource and Virtual Function Status Monitoring in Network Function Virtualization Environment Journal of Physcs: Conference Seres PAPER OPEN ACCESS Resource and Vrtual Functon Status Montorng n Network Functon Vrtualzaton Envronment To cte ths artcle: MS Ha et al 2018 J. Phys.: Conf. Ser. 1087

More information

An Optimal Bandwidth Allocation and Data Droppage Scheme for Differentiated Services in a Wireless Network

An Optimal Bandwidth Allocation and Data Droppage Scheme for Differentiated Services in a Wireless Network Purdue Unversty Purdue e-pubs ECE Techncal Reports Electrcal and Computer Engneerng 3--7 An Optmal Bandwdth Allocaton and Data Droppage Scheme for Dfferentated Servces n a Wreless Network Waseem Shekh

More information

Dynamic Bandwidth Allocation Schemes in Hybrid TDM/WDM Passive Optical Networks

Dynamic Bandwidth Allocation Schemes in Hybrid TDM/WDM Passive Optical Networks Dynamc Bandwdth Allocaton Schemes n Hybrd TDM/WDM Passve Optcal Networks Ahmad R. Dhan, Chad M. Ass, and Abdallah Sham Concorda Insttue for Informaton Systems Engneerng Concorda Unversty, Montreal, Quebec,

More information

Enhancing Class-Based Service Architectures with Adaptive Rate Allocation and Dropping Mechanisms

Enhancing Class-Based Service Architectures with Adaptive Rate Allocation and Dropping Mechanisms Enhancng Class-Based Servce Archtectures wth Adaptve Rate Allocaton and Droppng Mechansms Ncolas Chrstn, Member, IEEE, Jörg Lebeherr, Senor Member, IEEE, and Tarek Abdelzaher, Member, IEEE Abstract Class-based

More information

Support Vector Machines

Support Vector Machines /9/207 MIST.6060 Busness Intellgence and Data Mnng What are Support Vector Machnes? Support Vector Machnes Support Vector Machnes (SVMs) are supervsed learnng technques that analyze data and recognze patterns.

More information

Meta-heuristics for Multidimensional Knapsack Problems

Meta-heuristics for Multidimensional Knapsack Problems 2012 4th Internatonal Conference on Computer Research and Development IPCSIT vol.39 (2012) (2012) IACSIT Press, Sngapore Meta-heurstcs for Multdmensonal Knapsack Problems Zhbao Man + Computer Scence Department,

More information

Comparisons of Packet Scheduling Algorithms for Fair Service among Connections on the Internet

Comparisons of Packet Scheduling Algorithms for Fair Service among Connections on the Internet Comparsons of Packet Schedulng Algorthms for Far Servce among Connectons on the Internet Go Hasegawa, Takahro Matsuo, Masayuk Murata and Hdeo Myahara Department of Infomatcs and Mathematcal Scence Graduate

More information

A Model Based on Multi-agent for Dynamic Bandwidth Allocation in Networks Guang LU, Jian-Wen QI

A Model Based on Multi-agent for Dynamic Bandwidth Allocation in Networks Guang LU, Jian-Wen QI 216 Jont Internatonal Conference on Artfcal Intellgence and Computer Engneerng (AICE 216) and Internatonal Conference on etwork and Communcaton Securty (CS 216) ISB: 978-1-6595-362-5 A Model Based on Mult-agent

More information

CHAPTER 2 PROPOSED IMPROVED PARTICLE SWARM OPTIMIZATION

CHAPTER 2 PROPOSED IMPROVED PARTICLE SWARM OPTIMIZATION 24 CHAPTER 2 PROPOSED IMPROVED PARTICLE SWARM OPTIMIZATION The present chapter proposes an IPSO approach for multprocessor task schedulng problem wth two classfcatons, namely, statc ndependent tasks and

More information

A Quantitative Assured Forwarding Service

A Quantitative Assured Forwarding Service TO APPEAR IN PROCEEDINGS OF IEEE INFOCOM 00, c IEEE A Quanttatve Assured Forwardng Servce Ncolas Chrstn, Jörg Lebeherr, and Tarek F. Abdelzaher Department of Computer Scence Unversty of Vrgna P.O. Box

More information

AADL : about scheduling analysis

AADL : about scheduling analysis AADL : about schedulng analyss Schedulng analyss, what s t? Embedded real-tme crtcal systems have temporal constrants to meet (e.g. deadlne). Many systems are bult wth operatng systems provdng multtaskng

More information

State of the Art in Differentiated

State of the Art in Differentiated Outlne Dfferentated Servces on the Internet Explct Allocaton of Best Effort Packet Delvery Servce, D. Clark and W. Fang A Two bt Dfferentated Servces Archtecture for the Internet, K. Nchols, V. Jacobson,

More information

Design and Experimental Evaluation of a Cross-Layer Deadline-Based Joint Routing and Spectrum Allocation Algorithm

Design and Experimental Evaluation of a Cross-Layer Deadline-Based Joint Routing and Spectrum Allocation Algorithm Desgn and Expermental Evaluaton of a Cross-Layer Deadlne-Based Jont Routng and Spectrum Allocaton Algorthm Jthn Jagannath, Sean Furman, Tommaso Meloda, and Andrew Drozd Abstract The desgn and mplementaton

More information

Gateway Algorithm for Fair Bandwidth Sharing

Gateway Algorithm for Fair Bandwidth Sharing Algorm for Far Bandwd Sharng We Y, Rupnder Makkar, Ioanns Lambadars Department of System and Computer Engneerng Carleton Unversty 5 Colonel By Dr., Ottawa, ON KS 5B6, Canada {wy, rup, oanns}@sce.carleton.ca

More information

On Achieving Fairness in the Joint Allocation of Buffer and Bandwidth Resources: Principles and Algorithms

On Achieving Fairness in the Joint Allocation of Buffer and Bandwidth Resources: Principles and Algorithms On Achevng Farness n the Jont Allocaton of Buffer and Bandwdth Resources: Prncples and Algorthms Yunka Zhou and Harsh Sethu (correspondng author) Abstract Farness n network traffc management can mprove

More information

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation

An Iterative Solution Approach to Process Plant Layout using Mixed Integer Optimisation 17 th European Symposum on Computer Aded Process Engneerng ESCAPE17 V. Plesu and P.S. Agach (Edtors) 2007 Elsever B.V. All rghts reserved. 1 An Iteratve Soluton Approach to Process Plant Layout usng Mxed

More information

Neural Network Control for TCP Network Congestion

Neural Network Control for TCP Network Congestion 5 Amercan Control Conference June 8-, 5. Portland, OR, USA FrA3. Neural Network Control for TCP Network Congeston Hyun C. Cho, M. Sam Fadal, Hyunjeong Lee Electrcal Engneerng/6, Unversty of Nevada, Reno,

More information

Game Based Virtual Bandwidth Allocation for Virtual Networks in Data Centers

Game Based Virtual Bandwidth Allocation for Virtual Networks in Data Centers Avaable onlne at www.scencedrect.com Proceda Engneerng 23 (20) 780 785 Power Electroncs and Engneerng Applcaton, 20 Game Based Vrtual Bandwdth Allocaton for Vrtual Networks n Data Centers Cu-rong Wang,

More information

Session 5.3. Switching/Routing and Transmission planning

Session 5.3. Switching/Routing and Transmission planning ITU Regonal Semnar Belgrade Serba and Montenegro 20-24 24 June 2005 Sesson 5.3 Swtchng/Routng and Transmsson plannng volvng nfrastructures to NGN and related Plannng Strateges and Tools I.S. Sesson 5.3-1

More information

Channel 0. Channel 1 Channel 2. Channel 3 Channel 4. Channel 5 Channel 6 Channel 7

Channel 0. Channel 1 Channel 2. Channel 3 Channel 4. Channel 5 Channel 6 Channel 7 Optmzed Regonal Cachng for On-Demand Data Delvery Derek L. Eager Mchael C. Ferrs Mary K. Vernon Unversty of Saskatchewan Unversty of Wsconsn Madson Saskatoon, SK Canada S7N 5A9 Madson, WI 5376 eager@cs.usask.ca

More information

Feature Reduction and Selection

Feature Reduction and Selection Feature Reducton and Selecton Dr. Shuang LIANG School of Software Engneerng TongJ Unversty Fall, 2012 Today s Topcs Introducton Problems of Dmensonalty Feature Reducton Statstc methods Prncpal Components

More information

Adaptive Resource Allocation Control with On-Line Search for Fair QoS Level

Adaptive Resource Allocation Control with On-Line Search for Fair QoS Level Adaptve Resource Allocaton Control wth On-Lne Search for Far QoS Level Fumko Harada, Toshmtsu Usho, Graduate School of Engneerng Scence Osaka Unversty {harada@hopf, usho@}sysesosaka-uacjp Yukkazu akamoto

More information

S1 Note. Basis functions.

S1 Note. Basis functions. S1 Note. Bass functons. Contents Types of bass functons...1 The Fourer bass...2 B-splne bass...3 Power and type I error rates wth dfferent numbers of bass functons...4 Table S1. Smulaton results of type

More information

Design of the Application-Level Protocol for Synchronized Multimedia Sessions

Design of the Application-Level Protocol for Synchronized Multimedia Sessions Desgn of the Applcaton-Level Protocol for Synchronzed Multmeda Sessons Chun-Chuan Yang Multmeda and Communcatons Laboratory Department of Computer Scence and Informaton Engneerng Natonal Ch Nan Unversty,

More information

Performance Analysis of a Managed Bandwidth Service for ATM Networks

Performance Analysis of a Managed Bandwidth Service for ATM Networks Performance Analyss of a Managed Bandwdth Servce for ATM Networks Chrstos Bouras Chryss Chantz Vaggels Kapoulas Alexandros Panagopoulos Ioanna Sampraku Afrodte Sevast Research Academc Computer Technology

More information

3. CR parameters and Multi-Objective Fitness Function

3. CR parameters and Multi-Objective Fitness Function 3 CR parameters and Mult-objectve Ftness Functon 41 3. CR parameters and Mult-Objectve Ftness Functon 3.1. Introducton Cogntve rados dynamcally confgure the wreless communcaton system, whch takes beneft

More information

THere are increasing interests and use of mobile ad hoc

THere are increasing interests and use of mobile ad hoc 1 Adaptve Schedulng n MIMO-based Heterogeneous Ad hoc Networks Shan Chu, Xn Wang Member, IEEE, and Yuanyuan Yang Fellow, IEEE. Abstract The demands for data rate and transmsson relablty constantly ncrease

More information

Reducing Frame Rate for Object Tracking

Reducing Frame Rate for Object Tracking Reducng Frame Rate for Object Trackng Pavel Korshunov 1 and We Tsang Oo 2 1 Natonal Unversty of Sngapore, Sngapore 11977, pavelkor@comp.nus.edu.sg 2 Natonal Unversty of Sngapore, Sngapore 11977, oowt@comp.nus.edu.sg

More information

Regional Load Balancing Circuitous Bandwidth Allocation Method Based on Dynamic Auction Mechanism

Regional Load Balancing Circuitous Bandwidth Allocation Method Based on Dynamic Auction Mechanism ATEC Web of Conferences 76, (8) IFID 8 https://do.org/./matecconf/876 Regonal Load Balancng Crcutous Bandwdth Allocaton ethod Based on Dynamc Aucton echansm Wang Chao, Zhang Dalong, Ran Xaomn atonal Dgtal

More information

Delay Variation Optimized Traffic Allocation Based on Network Calculus for Multi-path Routing in Wireless Mesh Networks

Delay Variation Optimized Traffic Allocation Based on Network Calculus for Multi-path Routing in Wireless Mesh Networks Appl. Math. Inf. Sc. 7, No. 2L, 467-474 2013) 467 Appled Mathematcs & Informaton Scences An Internatonal Journal http://dx.do.org/10.12785/ams/072l13 Delay Varaton Optmzed Traffc Allocaton Based on Network

More information

Routing in Degree-constrained FSO Mesh Networks

Routing in Degree-constrained FSO Mesh Networks Internatonal Journal of Hybrd Informaton Technology Vol., No., Aprl, 009 Routng n Degree-constraned FSO Mesh Networks Zpng Hu, Pramode Verma, and James Sluss Jr. School of Electrcal & Computer Engneerng

More information

TECHNICAL REPORT AN OPTIMAL DISTRIBUTED PROTOCOL FOR FAST CONVERGENCE TO MAXMIN RATE ALLOCATION. Jordi Ros and Wei K Tsai

TECHNICAL REPORT AN OPTIMAL DISTRIBUTED PROTOCOL FOR FAST CONVERGENCE TO MAXMIN RATE ALLOCATION. Jordi Ros and Wei K Tsai TECHNICAL REPORT AN OPTIMAL DISTRIUTED PROTOCOL FOR FAST CONVERGENCE TO MAXMIN RATE ALLOCATION Jord Ros and We K Tsa Department of Electrcal and Computer Engneerng Unversty of Calforna, Irvne 1 AN OPTIMAL

More information

Why Congestion Control. Congestion Control and Active Queue Management. TCP Congestion Control Behavior. Generic TCP CC Behavior: Additive Increase

Why Congestion Control. Congestion Control and Active Queue Management. TCP Congestion Control Behavior. Generic TCP CC Behavior: Additive Increase Congeston Control and Actve Queue Management Congeston Control, Effcency and Farness Analyss of TCP Congeston Control A smple TCP throughput formula RED and Actve Queue Management How RED works Flud model

More information

Fast Retransmission of Real-Time Traffic in HIPERLAN/2 Systems

Fast Retransmission of Real-Time Traffic in HIPERLAN/2 Systems Fast Retransmsson of Real-Tme Traffc n HIPERLAN/ Systems José A Afonso and Joaqum E Neves Department of Industral Electroncs Unversty of Mnho, Campus de Azurém 4800-058 Gumarães, Portugal {joseafonso,

More information

For instance, ; the five basic number-sets are increasingly more n A B & B A A = B (1)

For instance, ; the five basic number-sets are increasingly more n A B & B A A = B (1) Secton 1.2 Subsets and the Boolean operatons on sets If every element of the set A s an element of the set B, we say that A s a subset of B, or that A s contaned n B, or that B contans A, and we wrte A

More information

FORESIGHTED JOINT RESOURCE RECIPROCATION AND SCHEDULING STRATEGIES FOR REAL-TIME VIDEO STREAMING OVER PEER-TO-PEER NETWORKS

FORESIGHTED JOINT RESOURCE RECIPROCATION AND SCHEDULING STRATEGIES FOR REAL-TIME VIDEO STREAMING OVER PEER-TO-PEER NETWORKS FORESIGHTED JOINT RESOURCE RECIPROCATION AND SCHEDULING STRATEGIES FOR REAL-TIME VIDEO STREAMING OVER PEER-TO-PEER NETWORKS Sunghoon Ivan Lee, Hyunggon Park, and Mhaela van der Schaar Electrcal Engneerng

More information

Steps for Computing the Dissimilarity, Entropy, Herfindahl-Hirschman and. Accessibility (Gravity with Competition) Indices

Steps for Computing the Dissimilarity, Entropy, Herfindahl-Hirschman and. Accessibility (Gravity with Competition) Indices Steps for Computng the Dssmlarty, Entropy, Herfndahl-Hrschman and Accessblty (Gravty wth Competton) Indces I. Dssmlarty Index Measurement: The followng formula can be used to measure the evenness between

More information

Can Congestion Control and Traffic Engineering Be at Odds?

Can Congestion Control and Traffic Engineering Be at Odds? Can Congeston Control and Traffc Engneerng Be at Odds? Jayue He Dept. of EE, Prnceton Unversty Emal: jhe@prnceton.edu Mung Chang Dept. of EE, Prnceton Unversty Emal: changm@prnceton.edu Jennfer Rexford

More information

User Authentication Based On Behavioral Mouse Dynamics Biometrics

User Authentication Based On Behavioral Mouse Dynamics Biometrics User Authentcaton Based On Behavoral Mouse Dynamcs Bometrcs Chee-Hyung Yoon Danel Donghyun Km Department of Computer Scence Department of Computer Scence Stanford Unversty Stanford Unversty Stanford, CA

More information

Virtual Memory. Background. No. 10. Virtual Memory: concept. Logical Memory Space (review) Demand Paging(1) Virtual Memory

Virtual Memory. Background. No. 10. Virtual Memory: concept. Logical Memory Space (review) Demand Paging(1) Virtual Memory Background EECS. Operatng System Fundamentals No. Vrtual Memory Prof. Hu Jang Department of Electrcal Engneerng and Computer Scence, York Unversty Memory-management methods normally requres the entre process

More information

QoS-aware routing for heterogeneous layered unicast transmissions in wireless mesh networks with cooperative network coding

QoS-aware routing for heterogeneous layered unicast transmissions in wireless mesh networks with cooperative network coding Tarno et al. EURASIP Journal on Wreless Communcatons and Networkng 214, 214:81 http://wcn.euraspournals.com/content/214/1/81 RESEARCH Open Access QoS-aware routng for heterogeneous layered uncast transmssons

More information

5 The Primal-Dual Method

5 The Primal-Dual Method 5 The Prmal-Dual Method Orgnally desgned as a method for solvng lnear programs, where t reduces weghted optmzaton problems to smpler combnatoral ones, the prmal-dual method (PDM) has receved much attenton

More information

Solutions for Real-Time Communication over Best-Effort Networks

Solutions for Real-Time Communication over Best-Effort Networks Solutons for Real-Tme Communcaton over Best-Effort Networks Anca Hangan, Ramona Marfevc, Gheorghe Sebestyen Techncal Unversty of Cluj-Napoca, Computer Scence Department {Anca.Hangan, Ramona.Marfevc, Gheorghe.Sebestyen}@cs.utcluj.ro

More information

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points;

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points; Subspace clusterng Clusterng Fundamental to all clusterng technques s the choce of dstance measure between data ponts; D q ( ) ( ) 2 x x = x x, j k = 1 k jk Squared Eucldean dstance Assumpton: All features

More information

Fibre-Optic AWG-based Real-Time Networks

Fibre-Optic AWG-based Real-Time Networks Fbre-Optc AWG-based Real-Tme Networks Krstna Kunert, Annette Böhm, Magnus Jonsson, School of Informaton Scence, Computer and Electrcal Engneerng, Halmstad Unversty {Magnus.Jonsson, Krstna.Kunert}@de.hh.se

More information

WITH rapid improvements of wireless technologies,

WITH rapid improvements of wireless technologies, JOURNAL OF SYSTEMS ARCHITECTURE, SPECIAL ISSUE: HIGHLY-RELIABLE CPS, VOL. 00, NO. 0, MONTH 013 1 Adaptve GTS Allocaton n IEEE 80.15.4 for Real-Tme Wreless Sensor Networks Feng Xa, Ruonan Hao, Je L, Naxue

More information

Biostatistics 615/815

Biostatistics 615/815 The E-M Algorthm Bostatstcs 615/815 Lecture 17 Last Lecture: The Smplex Method General method for optmzaton Makes few assumptons about functon Crawls towards mnmum Some recommendatons Multple startng ponts

More information