Bid-Price Control for Energy-Aware Pricing of Cloud Services

Size: px
Start display at page:

Download "Bid-Price Control for Energy-Aware Pricing of Cloud Services"


1 Bid-Price Conrol for Energy-Aware Pricing of Cloud Services Marc Premm Universiy of Hohenheim, Deparmen of Informaion Sysems 2, Sugar, Germany Absrac. The amoun of elecrical energy consumed by Cloud compuing resources keeps rising coninuously. To exploi he full poenial of reducing he carbon fooprin, echnical opimizaion of daa cener load and cooling disribuion is no sufficien. We propose a mehod ha moivaes Cloud service providers o inves in energy-efficien infrasrucure, which hen allows for increasing revenue. The differeniaion beween convenional and green services offers he possibiliy o apply price discriminaion approaches known from Revenue Managemen lieraure. Applying bid-price conrolled pricing for he provider s decision on acceping an incoming reques bears he poenial of increased revenue. We demonsrae he efficacy of he developed arifac hrough an experimenal evaluaion for various seings of supply and demand. Keywords: Cloud Compuing, Revenue Managemen, Energy-efficiency, Bid- Price Conrol 1 Inroducion Since he amoun of IT devices and heir elecrical energy consumpion are coninuously increasing [1], a lo of effors dedicaed o GreenIT have been carried ou o reduce he carbon fooprin [2]. Energy-aware compuaion has spread furher since he adven of Cloud compuing wih is daa ceners epiomizing he huge amoun of energy consumed by oday s IT. The lieraure already provides mehods o increase energy efficiency on a echnical level [3]. From an economic perspecive, however, here is addiional poenial of reducing he carbon fooprin by bringing he opimal se of conracors ogeher. Business praciioners are beginning o realize he imporance of susainabiliy [4] and recen sudies show srong evidence ha he willingness o pay for green labeled producs is higher han for convenional ones [5]. An increasing number of Cloud service providers, e.g. and GreenQloud, are propagaing he carbon free resource consumpion of heir offered services. Various levels of willingness o pay for conracually guaraneed energy-efficien Cloud service provision bear poenials o increase he providers revenues by applying price discriminaion mehods. As hese Cloud service providers don reveal heir inernal pricing sraegy, we found no evidence ha his green aspec is also used for price discriminaion. Green resources 12 h Inernaional Conference on Wirschafsinformaik, March , Osnabrück, Germany

2 allow Cloud service providers o offer green services, bu may also hos services for which no special energy clause has been included in he service level agreemen. For he Cloud service provider, his versailiy leads o he decision problem wheher o accep an incoming reques ha has o be hosed on green resources, hough i wouldn be necessary from a conracual perspecive. Each new conrac binds resources also in fuure ime periods, and hus may hinder fuure requess wih a higher willingness o pay for green services. Fig. 1 shows a brief example of wo poenial service consumers ha reques services from a provider, boh wih differen prioriizaions on he energy-efficiency. The provider on he oher hand has wo differen kinds of resources a disposal o deliver conraced services. For modeling and simulaing he disribued naure of his scenario, approaches from muliagen research are suiable [6]. Drawing from Revenue Managemen (RM) lieraure, we propose a mehod for he decision over acceping an incoming reques wih he aim o maximize revenue as an exension of exising pricing mechanisms. Previous work has shown ha he principles of RM can be purposefully adaped o model Cloud compuing scenarios [7]. We exen he proposed model by including wo ypes of resources and by supporing long-erm requess wih an apriori unknown conrac lengh. In conras o oher approaches ha require a rused hird pary, he mehod can be implemened by a provider wih differen kinds of resources wihou changing he way of propagaing is services, e.g. he usage of green or convenional resources may be promoed by differen brands of he same corporaion. Hence, pricediscriminaion based on energy-efficiency parameers is no necessarily observable by he consumer and, hus, avoids he risk of negaive side-effecs. We show ha rejecing requess for convenional services in cerain demand siuaions increases he provider s revenue and, hus, he mehod moivaes providers o rely on energy-efficien service provision from an economical perspecive wih he consequence of an increased ecological susainabiliy. In summary, he objecives of his research are (1) o develop a mehod ha increases expeced revenues hrough price discriminaion based on energy-efficiency and (2) o demonsrae he efficacy of he mehod hrough an experimenal evaluaion. The remainder of he paper is organized as follows. Secion 2 discusses relaed work. Secion 3 inroduces he formal framework. The proposed mehod is presened in Secion 4 and evaluaed in Secion 5. Secion 6 concludes. Cloud service provider Service consumer 1 Green resources Convenional resources Service consumer 2 Fig. 1. Service requess for differen kinds of energy efficiency

3 2 Theoreical Background The heoreical background is hree-fold: (i) Secion 2.1 discusses relaed work on energy-ware pricing for Cloud compuing, (ii) due o he concepual similariies o Cloud scenarios, Secion 2.2 inroduces he paradigm of muliagen organizaions for sofware agens, and (iii) Secion 2.3 paves he way for he applicaion of mehods from RM. 2.1 Energy-Aware Pricing for Cloud Compuing We use he erm Cloud compuing in accordance wih he NIST-definiion [8] and require a Cloud service o saisify he following crieria: (i) auomaizaion, (ii) sandardizaion, and (iii) scalabiliy. A disincion beween Infrasrucure as a Service (IaaS), Plaform as a Service (PaaS), and Sofware as a Service (SaaS) is no necessarily required for a provider ha hoss he whole infrasrucure, bu may be relevan for a SaaS provider ha needs o ren green infrasrucure o offer energy-efficien services o is cusomers. The erms green and energy-efficien are used synonymously hroughou his work, and as he decision problem iself is independen from he various available merics [9], we do no define or chose any. The expecaions on Cloud compuing are differen for he main wo groups of acors. Cusomers aim a moving fixed coss o variable coss and gain he abiliy of nearly unlimied resource scaling. This change in cos srucure is achieved hrough shor-erm conracs wih a ypical duraion of a leas one hour. Providers on he oher side focus on lower overall prices hrough sandardizaion and auomaizaion, i.e. here is a sandardized se of offered services which are provisioned in an auomaed fashion. Mos exising approaches relaed o energy-efficiency in Cloud compuing aim a reducing oal energy consumpion by opimizing load and cooling disribuion wihin a single daa cenre [10], [11]. However, some already consider he conrac negoiaion phase as an essenial sep owards susainabiliy. Garg e al. [12] sugges inroducing a Cloud broker ha behaves on behalf of he service consumer and selecs he mos energy-efficien Cloud provider for each reques. However, he use of a rused hird pary is an obsacle for he successful implemenaion, as no every consumer is willing o give up sovereigny. An approach ha can be applied o nearly any Cloud service scenario is provided by Haque e al. [13], who propose Green Service Level Agreemens (SLA) wih a quanifiable green rae. The proposed infrasrucure enables he differeniaion of green and convenional services based on elecrical energy in a single daacener, while a service provider has he possibiliy o accep or rejec a reques depending on he requesed service level. The auhors show ha a price discriminaing approach based on energy-efficiency promises increased revenues, bu lack a mehod o handle an oversupply of green resources.

4 2.2 Muliagen Organizaions A muliagen sysem (MAS) serves as a plaform for auonomous sofware agens ha allow represening scenarios wih disribued auhoriy and, hence, fis ino he conex of Cloud service providers and consumers [14]. We consider auonomous sofware agens ha are able o ac and reac wihin a MAS depuizing a naural person or anoher legal eniy. There are concepional similariies beween he negoiaions beween a provider and a consumer, as well as beween sofware agens wihin a MAS, also providing services o each oher. The provider independen TOSCA sandard (Topology and Orchesraion Specificaion for Cloud Applicaions) enables he applicaion of sofware agens ha may auonomously search and conrac Cloud services. Even if none of he acors in he Cloud compuing conex would acually apply agen-echnologies, we may draw from MAS lieraure for he developmen of our decision model. In a business conex, however, i is no always suiable o have a oally flexible MAS, where sofware agens are only loosely coupled. Insead, research in he area of MAS akes he direcion of sabilizing srucures wihin MAS drawing from organizaional heory. There are wo main differences in he conex and approaches in lieraure: (a) The aim of he organizaional srucure wihin he MAS is o represen he organizaional srucure of a social sysem wih clearly defined boundaries [15], [16], e.g. a company wih employees ha are represened by a sofware agen, or (b) approaches from social science are used o adap he behavior of sofware agens [17], e.g. o form coaliions or groups poenially across company borders. Horling and Lesser [17] discuss he differences beween differen erms relaed o muliagen organizaions, namely he mos widespread ones coaliions [18], [19], eams [18], [20] and congregaions [21]. They all have in common ha sofware agens form some kind of organizaional srucure allowing an improved cooperaion beween he paricipans. We define a muliagen organizaion as a long erm orienaed srucure ha compensaes sofware agens for providing services o he organizaion. Thus, we denoe all sofware agens ha have a conracual binding wih a muliagen organizaion o provide services as is paricipans. However, we do no require ha he goals or any subse of he goals of he muliagen organizaion and hose of is paricipans are necessarily he same. From he perspecive of a single auonomous sofware agen a decision has o be made wheher he paricipaion in a cerain muliagen organizaion is expeced o be beneficial for he agen. In a Cloud compuing conex his corresponds o he decision of a provider respecively a consumer wheher an offer of he counerpar will be acceped. For muliagen sysems Kraus [22] suggess o adap mehods from game heory and Operaions Research. While game heoreic conceps are broadly used for muliagen sysems, here are fields of Operaions Research ha are paid much less aenion. The nex secion presens he background on Revenue Managemen as par of Operaions Research and is applicaion o muliagen organizaions in he conex of energy-aware Cloud compuing.

5 2.3 Revenue Managemen RM has is origins in he airline indusry, where due o regulaions in he lae 1970s, companies were searching for new ways o increase revenues, e.g. by overbooking exising resources [23]. Since hen, RM has been exensively developed and has a firm place in pricing of various oher goods and services, including hoel rooms as well as renal cars. All applicaions of RM share some common characerisics: The provided good or services is (i) perishable, (ii) can be booked in advance and (iii) resource are scarce. The base example in an airline scenario is one scheduled fligh. Empy seas on he fligh canno be sored for laer use, so he revenue for his service is los (i). Cusomers are able o book he fligh in advance (ii), bu he oally available amoun of seas on he airplane is limied (iii). RM now aims a maximizing revenue by exending he se of poenial cusomers wih price discriminaion, e.g. offering he same fligh in various booking classes wih differen cancellaion policies. However, he main challenge arises wih he risk of selling oo many low priced services and, hus, he need for an appropriae procedure o selec he offered booking classes. An approach o enhance pricing in he conex of formaion and exension of muliagen organizaions has been par of previous work [24]. The mehod uses principles of RM aiming a opimizing profis and, hus, uiliy for each agen applying he mehod. Anoher previous work [6] addresses he applicaion of RM for Cloud services. Boh mehods assume ha he requesed Cloud service or organizaional membership is se for a specific ime horizon. However, in many Cloud compuing scenarios a Cloud service usually is requesed wihou a long lead ime and doesn have a specified end poin a he ime requesing he service. Insead, a Cloud service is generally requesed immediaely and conraced for an uncerain amoun of ime. Anoher requiremen for he applicaion of RM mehods is he presence of a possible discriminaion crierion. For a Cloud service, nearly all aspecs of a SLA may serve as a discriminaion crierion [6]. However, pricing according o energy-relaed variables are already common for some indusries like logisics. A higher willingness o pay of a cerain se of cusomers for green services offers he possibiliy o skim off addiional profi and, hus, ses he basis for he applicaion of RM mehods [13]. Hence, we draw from lieraure on RM, o design a pricing mehod for an energyaware Cloud compuing conex. 3 Formal Framework Convenional RM mehods are focused on a cerain ype of produc or service, like one single fligh or he provision of a hoel room for one nigh. These producs or services have a fixed ime span beween he sar and he end. I is quie uncommon, if no enirely impossible, o book a hoel room for an unlimied ime. However, his marks a difference beween convenional RM applicaions and Cloud compuing: A Cloud service is usually requesed immediaely for an uncerain amoun of ime. The applicabiliy in general of RM mehods is no affeced by his difference as he main requiremens poined ou in Secion 2.3 are sill saisfied. However, models common-

6 ly used in RM mus be adaped o he energy-aware Cloud compuing conex. We rely on previous work connecing RM wih Cloud services [7] as well as muliagen organizaions [24], and adap he models o cover requess for an unlimied ime span. 3.1 Services and Resources We consider sofware agens ha represen Cloud providers and ha provide services o a muliagen organizaion while receiving compensaion in reurn. We assume ha he Cloud provider has a fixed se of services S = {s 1,,s N }, while he exac ype of service is no furher specified and, hus, each service migh be SaaS, PaaS or IaaS. The discree ime axis is represened by {0,,T}, while each represens a cerain ime span depending on he ype of service. For each service, he provider can guaranee a cerain minimum level of energyefficiency for he devices used for service provision. We draw from RM lieraure and, hus, call he differen caegories booking classes. I is also possible o combine energy-relaed discriminaion crieria wih oher conracual side condiions usually used in service level agreemens, e.g. availabiliy, response ime, ec. These addiional crieria could be combined o increase he discriminaing disance beween wo consecuive booking classes. Despie he general possibiliy for an arbirary number of discriminaion levels, for he sake of simpliciy, we only consider wo booking classes b B = {Green, Conv}, while Conv represens a convenional operaion wihou any energy-consideraions and Green he sole execuion on energy-efficien devices. Each booking class has a fixed price p s,b for each service s and booking class b ha is consan for all {0,,T}. There are no wihdrawal resricions, so every service consumer may choose o wihdraw anyime wihou a penaly. The service providing agen has o rely on resources o be able o fulfill is conracual obligaions. We assume ha here are wo ypes of resources: Convenional resources ha are only able o be used for booking class Conv as well as energyefficien resources ha comply o boh booking classes and, hence, are more flexible for he provider. The capaciy lef for boh resources a ime is represened by he vecor x Green x. xconv The concree ype of differeniaion beween he wo resource pools is no relevan for he presened mehod. There are numerous possible crieria o measure he energyefficiency of IT resources, e.g. PUE for a whole compuaion cener or SPECpower for a single server [9]. The expeced capaciy lef in for some fuure ime ˆ wih ˆ T is represened by he funcion E xb, ˆ. We assume ha each service consumes a cerain amoun of resources independenly of he chosen booking class. The vecor m1 m m N

7 expresses how much resources each service consumes. One service s 1 in booking class b = Green will, hus, reduce he capaciy lef x Green by sm. 1 1 In booking class b = Conv he same service would be able o reduce x Green or x Conv by he same amoun. 3.2 Requess The value nework of a service consumer requesing a Cloud service may be referenced o a muliagen organizaion ha requess services offered by anoher agen. The service can be requesed in boh possible booking classes, while he amoun of requesed services is represened by he vecor r 1 r. r N In accordance wih RM lieraure, we assume ha here is a mos one reques per ime. However, his resricion does no limi he applicabiliy as he ime span beween wo ime poins can be arbirary small. The accepance of an incoming reques will reduce he available resources of he service provider and, hence, limi his scope for fuure usage. The vecor c serves as an indicaor wheher and wha kind of a conrac has been signed in wih c = (1,0) for green services, respecively c = (0,1) for convenional services and c = (0,0) if no conrac has been signed. The capaciy lef for he nex ime period is recursively calculaed by x 1 x Green 0 c mr x c mr xconv x 1 Green x Conv oherwise. 0 0 mr (1) The second alernaive in Equaion (1) formalizes he possibiliy of he service provider o use green resources also o cover convenional service conracs. The oal capaciy of he provider is given in = 0 wih x 0. 4 Pricing Mechanism This secion presens he pricing mechanism ha allows a sofware agen represening a Cloud service provider o increase is revenue by improving he decision wheher o accep or decline an incoming reques. We propose a bid-price approach ha calculaes some sor of reservaion prices for each resource and ha does no affec he commonly used process of requesing a service and a response wihin a narrow ime frame.

8 4.1 Decision Model The decision ha has o be made is wheher o accep an incoming reques or o rejec i. Mos Cloud service providers will accep any incoming reques for a fixed price level as long as resources are available o serve he reques. However, he differeniaion in wo differen booking classes as presened in he preceding secion enables more complex decision models wih he aim o increase revenue. Bid-price conrol from RM lieraure uses a reservaion price for each of he resources o conrol booking class availabiliy. If convenional resources are available ha canno be used for providing green services, a reques for he corresponding booking class should always been acceped. The more advanced green resources, however, migh be used o provide convenional or green services and, hus, are he focus of he decision ha has o be made. The bid-price π a ime represens he reservaion price for green resources under which i is expeced o be subopimal o sell he lower priced convenional booking class. For an incoming requess ha involves he usage of green resources, he offering condiion a ime for booking Class Conv is N mr rp (2), s s Conv s0 where he vecor muliplicaion mr maps he requesed services o he amoun of required resources and he righ par of he inequaion represens he price for he incoming reques in he convenional booking class. 4.2 Decision Mechanism The decision abou refusing a reques even hough resources would have been available has a huge difference compared o radiional RM applicaions: The resources saved emporally by a refused fligh reques, migh be used o provide higher priced booking classes laer on. Of course, here is he risk ha he resources remain unused, bu in he presen case, he rejecion of a reques has he consequence of immediae revenue loss. This revenue has o be gained laer on, and his is only possible wih higher priced booking classes. For a differeniable valuaion funcion V (x ) he bid-price is defined as he parial differeniaion wih respec o he relevan resource variable. As he only relevan resource for he decision in ime is x Green, we can adap Lilewood s rule [25] for he presen case wih long-erm conracs and immediae service provision: ˆ x : V p P D E x, 1 Green Green Green xgreen (3) wih he average price per resource p b, he expeced demand D b (Δ) and he lower limi of he expeced capaciy lef based on he curren conracs ˆ E xb, each for a fuure ime span Δ and booking class Green.

9 4.3 Forecasing Sraegy The forecasing sraegy is based on he ideas of Lilewood [25], bu has o be fundamenally adaped for he presen case as wihdrawals may occur coninuously. The decision mechanism presened in he previous secion is mainly deermined by he qualiy of he demand and wihdrawal forecasing, as Equaion (3) is dependen on hese wo funcions. The probabiliy ha he demand for green services will be higher han he expeced resources lef can provide is seled by muliple variables: (i) he disribuion of boh funcions, (ii) heir parameers, and (iii) he considered ime span Δ. We assume ha he expeced wihdrawals follow an exponenial disribuion wih expeced mean conrac lengh b in. As he acual conrac lengh is no previously known, he expeced mean conrac lengh b is deermined by he average conrac lengh of already erminaed conracs. Wih he assumpion ha hese pas evens can be used o forecas fuure wihdrawal evens, we can reformulae E, ˆ x in for all upcoming periods ˆ o ˆ b, ˆ 1 E x x e. As saed before, he rejecion of an incoming reques for a lower priced booking class can only be beneficial for he Cloud service provider if he manages o creae enough revenue wih fuure higher priced conracs insead. Wih an expeced conrac lengh for green services of Green and a price relaion beween green and convenional booking classes of he ime frame in which a higher priced conrac should be signed pconv p Green pconv is se o 1 Green p. If he provider has o wai longer han Δ for a higher Green priced reques, he ime span is no long enough o gain more revenue despie he higher price for green services. The demand is expeced o be normally disribued wih mean μ b and sandard deviaion σ b for booking class b. Boh variables can be deermined analogously o he mean conrac lengh b by he observaion of pas requess. For he calculaion of π, he relevan variables μ Green and σ Green for booking class Green are observed for each ime period and he convoluion of muliple independen disribuions has o be calculaed. Wih he assumpion ha requess are equally disribued among he involved number of service providers N Provider, he convoluion of he demand disribuion funcions for he ime inerval Δ has he following wo parameers: ˆ N Green Provider ˆ. Green Wih he simplificaion ha he capaciy lef in he inerval Δ is regarded as he mean of E xgreen, ˆ in he corresponding inerval, he probabiliy ha he expeced demand in ˆ exceeds he expeced capaciy lef can be calculaed by

10 E xgreen,ˆ ˆ 2 1x ˆ 2 ˆ ˆ 1 P DGreen E xgreen, 1 e dx. ˆ 2 (4) Wih Equaions (3) and (4), he bid-price π can be calculaed in each ime inerval and provide he inpu for Equaion (2) o decide wheher an incoming reques will be acceped or rejeced. 5 Evaluaion This secion presens an experimenal evaluaion of he proposed arifac. We describe he experimenal seup, repor he simulaion resuls, and discuss he findings. 5.1 Experimenal Seup For he simulaion of he Cloud compuing scenario, we used he agen-based simulaion framework Repas Simphony [26]. Service consumer agens sen service requess o service providers, which may have made an offer depending on heir decision model. There were wo ypes of service providers: (1) A regular provider ha offered services if he remaining capaciy allowed i and (2) a bid-price provider ha used he proposed mehod and ha beside capaciy resricions also involved he forecasing of fuure requess. We simulaed a compeiive duopoly Cloud service marke wih one regular and one bid-price provider. To minimize he amoun of relevan variables ha influence resuls, providers only offered one ype of service ha exacly used one ype of resource when conraced implying m 1 = 1 for s 1. As only he more versaile green resources are of ineres o price discriminaion purposes, we considered one energy efficien compuaion cenre 0 0 and se xconv 0 respecively xgreen 1,000. The planning horizon of he provider agens was T = 5,000. The service consumer agens requesed he service in one of he wo booking classes and chose he cheapes proposal respecively chose randomly when receiving wo or more equivalen proposals. Service consumers of ype C1 requesed only lower priced convenional services while hose of ype C2 requesed green services. Demand had been generaed wih he following parameers for he probabiliy disribuion: Every consumer agen generaed one reques in every ime period wih a probabiliy of 10 percen. The amoun of services requesed was normally disribued wih mean 10 and sandard deviaion 5. The conrac lengh ha was no apriori known o he providers was normally disribued wih mean 1,000 and sandard deviaion 500.

11 The price for one uni of service s 1 in booking class Green was se o 10 moneary unis per ime period while he one for booking class Conv was reduced by 5 % wih respec o he firs one. 5.2 Resuls The simulaion was performed 100 imes and for each run he values of he ime period beween icks 1,000 and 3,000 have been used for evaluaion o sor ou fading in and ou effecs. Average revenue per ick was he firs value we invesigaed. Fig. 2 shows he percenage of average revenue ha he bid-price provider exceeded he average revenue of he regular provider depending on he demand seing. The number of boh ypes of service consumer agens has been varied on an exponenial scale from 1 o 1,024. Negaive values represen a surplus of he regular provider and only occured if requess for he green booking class were very rarely. In all oher seings he bid-price provider gained a surplus over he regular provider. The highes surplus of 3.91% has been gained wih 1,024 consumers of ype C1 and 64 of ype C2. Fig. 3 shows he percenage of he sandard deviaion of he periodically gained revenue in relaion wih he mean value. The bid-price approach had a significanly higher sandard deviaion of revenue. While in he firs scenario demand heighs and conrac lengh were generaed using a normal disribuion, we examined a second scenario where hose wo variables have been generaed using a Poisson disribuion wih he same mean parameers. Hence, he expeced disribuion by he bid-price provider and he acual one differed and a lower performance of he proposed approach was expeced. However, he resuls showed he same srucure as Fig. 2, given Poisson disribued demand variables and even exceeded he opimal seing for he firs scenario wih a surplus of 4.40%. Fig. 2. Surplus of bid-price provider in relaion o regular provider wih respec o revenue

12 0.14% 0.14% 0.12% 0.12% 0.10% 0.10% 0.08% 0.08% 0.06% 0.06% 0.04% 0.04% 0.02% 0.02% 0.00% 0.00% BP regular BP regular C C2 Fig. 3. Sandard deviaion in relaion o mean for C1 = % 2.5% 2.0% 2.0% 1.5% 1.5% 1.0% 1.0% 0.5% 0.5% 0.0% 0.0% Average conrac lengh Average conrac lengh Fig. 4. Average revenue surplus of bid-price provider depending on average conrac lengh We also examined he impac of he relaion of conrac lengh and he amoun of services per reques on he average revenue of he provider. The providers faced 10 consumers of each ype and he average conrac lengh has been varied negaively correlaed o he amoun of requesed services o remain he average demand unchanged. Fig. 4 shows he percenage of average revenue of he bid-price provider exceeding he average revenue of he regular one. 5.3 Discussion When looking a he resuls of he previous secion, one has o keep in mind ha he price difference beween he wo booking classes was only 5%. This difference marks an upper limi for he proposed mehod and is advance is no able o exceed his limi. Fig. 2 shows ha he advanage of he approach is srongly dependen on he relaion beween consumers wih a higher and hose wih a lower willingness o pay. The highes surplus was gained when he amoun of C1 consumers was high and he amoun of C2 consumers was above a cerain limi, approximaely 8. In his case, mos requess sen o he providers aim a he convenional booking class and blocked he resources of he regular provider, while he bid-price approach denied mos of hose requess. When addiionally he number of C2 consumers increased, he probabiliy ha a reques achieved a higher revenue increased analogously and, hence, reduced he advanage of he proposed mehod. Even in non-opimal seings he bidprice approach achieved a posiive surplus in nearly all cases.

13 Even wih a demand disribuion funcion ha he bid-price provider did no apriori anicipaed, he gained a significan surplus in many seings. The sandard deviaion of he periodical revenue was significanly higher, bu on an exremely low level wih respec o he mean value. The average revenue was of course highly dependen on he average conrac lengh. If all or mos conracs only las one period, here would be no need o reserve capaciy for fuure requess. The same holds for he number of requess per ime period, as a rejeced reques has o be compensaed by anoher higher priced reques wihin a relaively shor amoun of ime. The reasonable applicabiliy of he proposed mehod, hus, depends on he relaion beween conrac lengh and reques frequency. However, we did no observe significan drawbacks in seings ha do no mee hese requiremens. 6 Conclusion This paper presens a mehod o increase expeced revenues of a Cloud service provider using price discriminaion. We adaped and applied approaches from Revenue Managemen and used he varying willingness o pay for ecologically susainable services as he discriminaion crierion. The model successfully inroduced a rolling ime horizon ha is no common for Revenue Managemen mehods, bu necessary o represen real-world Cloud compuing scenarios. We evaluaed he proposed mehod in a muli-agen simulaion and showed is efficacy depending on differen seings of supply and demand. The mehod gains an advanage on regular approaches by rejecing lower priced requess when higher fuure revenue is expeced. The demand for an enhanced susainabiliy of corporae IT keeps increasing coninuously and Cloud service provider canno ignore his rend [4]. Our research has hree implicaions for pracice. Firs, he proposed mehod increases revenues for Cloud service providers ha are hosing green compuing resources. Second, he possibiliy of higher revenues addiionally moivaes providers o inves in more versaile energy-efficien infrasrucure fed wih carbon free produced elecrical energy and incorporaing price discriminaion possibiliies. Third, he individual opimizing of Cloud service providers in erms of revenue leads o a subsiuion of old inefficien resources by new versaile ones and, hus, a sociey-wide improvemen of ecological susainabiliy. The approach is limied o markes where demand exceeds supply as he scope is resriced o acceping or rejecing a reques, while here are no significan drawbacks in oher siuaions and may hus also be used in scenarios wih apriori unknown demand-supply raio. Anoher prerequisie is ha a new conrac reserves sufficien resources in relaion o he reques frequency, while oherwise a compensaion of a rejeced reques is rarely possible. The discriminaion crierion energy-efficiency may even be combined wih oher aspecs ha are usually par of SLAs and, hence, jusify a higher price difference han 5% wih a simulaneously increased revenue poenial. Exensions ha involve decommimen penaliies ino he negoiaion phase of Cloud service provision show major similariies o cancelaion respecively wihdraw-

14 al fees known from radiional RM applicaions and may furher enhance resuls [6]. The bid-price conrol approach and hus he proposed model is easily exendable and adapable, e.g. by adding addiional discriminaion levels. The proposed mehod is highly depending on he qualiy of demand forecasing, hence, advanced forecasing mehods [27] promise o furher increase revenue especially in non-opimal seings of supply and demand. Selling more services han he available resources may deliver is praciced in radiional RM domains as overbooking. Findings from RM indicae ha combining he proposed mehod wih overbooking posiively affecs revenue [28]. Acknowledgemens. This work has been suppored by (1) he projec MIGRATE!, funded by he German Federal Minisry of Economics and Technology (BMWi, FKZ 01ME11052), and (2) he ehealhmonior projec (hp://, funded by he European Commission under conrac FP References 1. Cook, G., Dowdall, T., Pomeranz, D., Wang, Y.: Clicking Green: How Companies are Creaing he Green Inerne. Greenpeace, Washingon (2014) 2. Murugesan, S.: Harnessing Green IT: Principles and Pracices. IT Pro 01-02, (2008) 3. Berl, A., Gelenbe, E., Girolamo, M., Giuliani, G., Meer, H., Dang, M.Q., Penikousis, K.: Energy-Efficien Cloud Compuing. The Compuer Journal, vol. 53, no. 7, (2010) 4. Brooks, S., Wang, X., Sarker, S.: Unpacking Green IS: A Review of he Exising Lieraure and Direcions for he Fuure. In: Brocke, J., Seidel, S., Recker, J. (Eds.), Green Business Process Managemen - Towards he Susainable Enerprise, pp Springer (2012) 5. Cronin, J.J., Smih, J.S., Gleim, M.R., Ramirez, E., Marinez, J.D.: Green markeing sraegies: an examinaion of sakeholders and he opporuniies hey presen. Journal of he Academy of Markeing Science 39, (2011) 6. An, B., Lesser, V., Irwin, D., Zink, M.: Auomaed Negoiaion wih Decommimen for Dynamic Resource Allocaion in Cloud Compuing. In: Proceedings of he 9 h Inernaional Conference on Auonomous Agens and Muliagen Sysems, pp , Torono (2010) 7. Anandasivam, A., Premm, M.: Bid Price Conrol and Dynamic Pricing in Clouds. In: Proceedings of he 17 h European Conference on Informaion Sysems, pp , Verona (2009) 8. Mell, P., Grance, T.: The NIST Definiion of Cloud Compuing. Naional Insiue of Sandards and Technology (2011) 9. Daim, T., Jusice, J., Krampis, M., Les, M., Subramanian, G., Thirumalai, M.: Daa cener merics An energy efficiency model for informaion echnology managers. Managemen of Environmenal Qualiy, vol. 20, no. 6, (2009) 10. Beloglazov, A., Abawajy, J., Buyya, R.: Energy-aware resource allocaion heurisics for efficien managemen of daa ceners for Cloud compuing. In: Fuure Generaion Compuer Sysems 28, (2012) 11. Baliga, J., Ayre, R.W.A., Hinon, K., Tucker, R.S.: Green Cloud Compuing: Balancing Energy in Processing Sorage and Transpor. Proceedings of he IEEE 99, (2010)

15 12. Garg, S.K., Yeo, C.S., Buyya, R.: Green Cloud Framework for Improving Carbon Efficiency of Clouds. In: Jeanno, E., Namys, R., Roman, J. (eds.) Euro-Par LNCS, vol. 6852, pp Springer, Heidelberg (2011) 13. Haque, M.E., Le, K., Gori, I., Bianchini, R., Nguyen, T.D.: Providing Green SLAs in High Performance Compuing Clouds. In: Proceedings of Inernaional Green Compuing Conference, Arlingon (2013) 14. Huhns, M. N., Singh, M. P. (Eds.): Research Direcions for Service-Oriened Muliagen Sysems. IEE Inerne Compuing Nov.-Dec., (2005) 15. Kirn, S., Gasser, L.: Organizaional Approaches o Coordinaion in Muli-agen Sysems. Journal i+i 4, (1998) 16. Hübner, J.F., Boissier, O., Kiio, R., Ricci, A.: Insrumening muli-agen organisaions wih organisaional arifacs and agens. Auonomous Agens and Muliagen Sysems 20, (2010) 17. Horling, B., Lesser, V.: A survey of muliagen organizaional paradigms. The Knowledge Engineering Review 19:4, (2004) 18. Wooldridge, M.: An Inroducion o MuliAgen Sysems (2nd Ed.). Chicheser: John Wiley & Sons (2009) 19. Sandholm, T. W.: Disribued Raional Decision Making. In: G. Weiss (Ed.), Muliagen Sysems - A Modern Approach o Disribued Arificial Inelligence, Chaper 5, pp MIT Press (1999) 20. Jennings, N. R.: Conrolling cooperaive problem solving in indusrial muli-agen sysems using join inenions. Arificial Inelligence 75, (1995) 21. Brooks, C., Durfee, E.: Congregaion formaion in muliagen sysems. Auonomous Agens and Muliagen Sysems 7 (1-2), (2003) 22. Kraus, S.: Negoiaion and Cooperaion in Muli-Agen Environmens. In: Arificial Inelligence 94, (1997) 23. Talluri, K.T., van Ryzin, G.J.: The Theory and Pracice of Revenue Managemen. Springer, New York (2004) 24. Premm, M., Widmer, T., Karaenke, P.: Bid-Price Conrol for he Formaion of Muliagen Organisaions. In: Klusch, M., Thimm, M., Paprzycki (eds.) MATE LNAI, vol. 8076, pp Springer, Heidelberg (2013) 25. Lilewood, K.: Forecasing and conrol of passenger bookings. In: Proceedings of he Twelfh Annual AGIFORS Symposium, Nahanya (1972) 26. Norh, M., Collier, N., Ozik, J., Taara, E., Macal, C., Bragen, M., Sydelko, P.: Complex adapive sysems modeling wih repas simphony. Complex Adapive Sysems Modeling 1(1) (2013) 27. Syneos, A.A., Boylan, J.E., Disney, S.M.: Forecasing for invenory planning: A 50-year review. The Journal of he Operaional Research Sociey 60, (2009) 28. Chiang, W.C., Chen, J., Xu, X.: An overview of research on revenue managemen: Curren issues and fuure research. Inernaional Journal of Revenue Managemen 1(1), (2007)

Towards Optimal Capacity Segmentation with Hybrid Cloud Pricing

Towards Optimal Capacity Segmentation with Hybrid Cloud Pricing Towards Opimal Capaciy Segmenaion wih Hybrid Cloud Pricing Wei Wang, Baochun Li, and Ben Liang Deparmen of Elecrical and Compuer Engineering Universiy of Torono Torono, ON M5S 3G4, Canada,

More information

2009 / 2 Review of Business and Economics. Federico Etro 1

2009 / 2 Review of Business and Economics. Federico Etro 1 The Economic Impac of Cloud Compuing on Business Creaion, Employmen and Oupu in Europe An applicaion of he Endogenous Marke Srucures Approach o a GPT innovaion Federico Ero ABSTRACT Cloud compuing is a

More information

On the Management of Life Insurance Company Risk by Strategic Choice of Product Mix, Investment Strategy and Surplus Appropriation Schemes

On the Management of Life Insurance Company Risk by Strategic Choice of Product Mix, Investment Strategy and Surplus Appropriation Schemes On he Managemen of Life Insurance Company Risk by raegic Choice of Produc Mix, Invesmen raegy and urplus Appropriaion chemes Alexander Bohner, Nadine Gazer, Peer Løche Jørgensen Working Paper Deparmen

More information

Optimal demand response: problem formulation and deterministic case

Optimal demand response: problem formulation and deterministic case Opimal demand response: problem formulaion and deerminisic case Lijun Chen, Na Li, Libin Jiang, and Seven H. Low Absrac We consider a se of users served by a single load-serving eniy (LSE. The LSE procures

More information

Basic Life Insurance Mathematics. Ragnar Norberg

Basic Life Insurance Mathematics. Ragnar Norberg Basic Life Insurance Mahemaics Ragnar Norberg Version: Sepember 22 Conens 1 Inroducion 5 1.1 Banking versus insurance...................... 5 1.2 Moraliy............................... 7 1.3 Banking................................

More information

Research. Michigan. Center. Retirement

Research. Michigan. Center. Retirement Michigan Universiy of Reiremen Research Cener Working Paper WP 2006-124 Opimizing he Reiremen Porfolio: Asse Allocaion, Annuiizaion, and Risk Aversion Wolfram J. Horneff, Raimond Maurer, Olivia S. Michell,

More information

A Working Solution to the Question of Nominal GDP Targeting

A Working Solution to the Question of Nominal GDP Targeting A Working Soluion o he Quesion of Nominal GDP Targeing Michael T. Belongia Oho Smih Professor of Economics Universiy of Mississippi Box 1848 Universiy, MS 38677 and Peer N. Ireland Deparmen

More information


EDUCATION POLICIES AND STRATEGIES EDUCATION POLICIES AND STRATEGIES Naional Educaion Secor Developmen Plan: A resul-based planning handbook 13 Educaion Policies and Sraegies 13 Educaion Policies and Sraegies 13 Naional Educaion Secor Developmen

More information

Cost-Sensitive Learning by Cost-Proportionate Example Weighting

Cost-Sensitive Learning by Cost-Proportionate Example Weighting Cos-Sensiive Learning by Cos-Proporionae Example Weighing Bianca Zadrozny, John Langford, Naoki Abe Mahemaical Sciences Deparmen IBM T. J. Wason Research Cener Yorkown Heighs, NY 0598 Absrac We propose

More information


NORTHWESTERN UNIVERSITY J.L. KELLOGG GRADUATE SCHOOL OF MANAGEMENT NORTHWESTERN UNIVERSITY J.L. KELLOGG GRADUATE SCHOOL OF MANAGEMENT Tim Thompson Finance D30 Teaching Noe: Projec Cash Flow Analysis Inroducion We have discussed applying he discouned cash flow framework

More information

Are Under- and Over-reaction the Same Matter? A Price Inertia based Account

Are Under- and Over-reaction the Same Matter? A Price Inertia based Account Are Under- and Over-reacion he Same Maer? A Price Ineria based Accoun Shengle Lin and Sephen Rasseni Economic Science Insiue, Chapman Universiy, Orange, CA 92866, USA Laes Version: Nov, 2008 Absrac. Theories

More information

Is China Over-Investing and Does it Matter?

Is China Over-Investing and Does it Matter? WP/12/277 Is China Over-Invesing and Does i Maer? Il Houng Lee, Muraza Syed, and Liu Xueyan 2012 Inernaional Moneary Fund WP/12/277 IMF Working Paper Asia and Pacific Deparmen Is China Over-Invesing and

More information

Today s managers are very interested in predicting the future purchasing patterns of their customers, which

Today s managers are very interested in predicting the future purchasing patterns of their customers, which Vol. 24, No. 2, Spring 25, pp. 275 284 issn 732-2399 eissn 1526-548X 5 242 275 informs doi 1.1287/mksc.14.98 25 INFORMS Couning Your Cusomers he Easy Way: An Alernaive o he Pareo/NBD Model Peer S. Fader

More information

Does Britain or the United States Have the Right Gasoline Tax?

Does Britain or the United States Have the Right Gasoline Tax? Does Briain or he Unied Saes Have he Righ Gasoline Tax? Ian W.H. Parry and Kenneh A. Small March 2002 (rev. Sep. 2004) Discussion Paper 02 12 rev. Resources for he uure 1616 P Sree, NW Washingon, D.C.

More information

Part-time Work, Wages and Productivity: Evidence from Matched Panel Data

Part-time Work, Wages and Productivity: Evidence from Matched Panel Data Par-ime Work, Wages and Produciviy: Evidence from Mached Panel Daa Alessandra Caaldi (Universià di Roma "La Sapienza" and SBS-EM) Sephan Kampelmann (Universié de Lille, CLERSE, SBS-EM) François Rycx (Universié

More information

Anchoring Bias in Consensus Forecasts and its Effect on Market Prices

Anchoring Bias in Consensus Forecasts and its Effect on Market Prices Finance and Economics Discussion Series Divisions of Research & Saisics and Moneary Affairs Federal Reserve Board, Washingon, D.C. Anchoring Bias in Consensus Forecass and is Effec on Marke Prices Sean

More information

EMBARGO: December 4th, 2014, 11am Pacific/2pm Eastern/7pm UK. The Social Bayesian Brain: Does Mentalizing Make a Difference When We Learn?

EMBARGO: December 4th, 2014, 11am Pacific/2pm Eastern/7pm UK. The Social Bayesian Brain: Does Mentalizing Make a Difference When We Learn? EMBARGO: December 4h, 2014, 11am Pacific/2pm Easern/7pm UK The Social Bayesian Brain: Does Menalizing Make a Difference When We Learn? Marie Devaine 1,2, Guillaume Hollard 3,4, Jean Daunizeau 1,2,5 * 1

More information

The U.S. Treasury Yield Curve: 1961 to the Present

The U.S. Treasury Yield Curve: 1961 to the Present Finance and Economics Discussion Series Divisions of Research & Saisics and Moneary Affairs Federal Reserve Board, Washingon, D.C. The U.S. Treasury Yield Curve: 1961 o he Presen Refe S. Gurkaynak, Brian

More information



More information


THE IMPACT OF WEATHER ON TRANSIT RIDERSHIP IN CHICAGO THE IMPACT OF WEATHER ON TRANSIT RIDERSHIP IN CHICAGO Zhan Guo, Massachuses Insiue of Technology, Room 9-524, 77 Massachuses Ave., Cambridge, MA, 02139, Nigel H.M. Wilson (Corresponding

More information

The Macroeconomics of Medium-Term Aid Scaling-Up Scenarios

The Macroeconomics of Medium-Term Aid Scaling-Up Scenarios WP//6 The Macroeconomics of Medium-Term Aid Scaling-Up Scenarios Andrew Berg, Jan Goschalk, Rafael Porillo, and Luis-Felipe Zanna 2 Inernaional Moneary Fund WP//6 IMF Working Paper Research Deparmen The

More information

Tax Deductible Spending, Environmental Policy, and the "Double Dividend" Hypothesis

Tax Deductible Spending, Environmental Policy, and the Double Dividend Hypothesis Tax Deducible Spending, Environmenal Policy, and he "Double Dividend" ypohesis Ian W.. Parry Anonio Miguel Beno Discussion Paper 99-24 February 999 66 P Sree, W Washingon, DC 20036 Telephone 202-328-5000

More information

Central Bank Communication: Different Strategies, Same Effectiveness?

Central Bank Communication: Different Strategies, Same Effectiveness? Cenral Bank Communicaion: Differen Sraegies, Same Effeciveness? Michael Ehrmann and Marcel Frazscher * European Cenral Bank, November 2004 Absrac The paper

More information

The concept of potential output plays a

The concept of potential output plays a Wha Do We Know (And No Know) Abou Poenial Oupu? Susano Basu and John G. Fernald Poenial oupu is an imporan concep in economics. Policymakers ofen use a one-secor neoclassical model o hink abou long-run

More information

Forecasting Exchange Rates Out-of-Sample with Panel Methods and Real-Time Data. Onur Ince * University of Houston

Forecasting Exchange Rates Out-of-Sample with Panel Methods and Real-Time Data. Onur Ince * University of Houston Forecasing Exchange Raes Ou-of-Sample wih anel Mehods and Real-ime Daa Onur Ince * Universiy of Houson Absrac his paper evaluaes ou-of-sample exchange rae forecasing wih urchasing ower ariy () and aylor

More information

Improved Techniques for Grid Mapping with Rao-Blackwellized Particle Filters

Improved Techniques for Grid Mapping with Rao-Blackwellized Particle Filters 1 Improved Techniques for Grid Mapping wih Rao-Blackwellized Paricle Filers Giorgio Grisei Cyrill Sachniss Wolfram Burgard Universiy of Freiburg, Dep. of Compuer Science, Georges-Köhler-Allee 79, D-79110

More information

-DR 03021 - Anne CAZAVAN-JENY* Thomas JEANJEAN** Juillet 2003

-DR 03021 - Anne CAZAVAN-JENY* Thomas JEANJEAN** Juillet 2003 CENTRE DE RECHERCHE RESEARCH CENTER -DR 03021 - Value Relevance of R&D Reporing : A Signaling Inerpreaion Anne CAZAVAN-JENY* & Thomas JEANJEAN** Juille 2003 * CAZAVAN-JENY A. ESSEC, Avenue B. Hirsch, BP

More information

When Should Public Debt Be Reduced?

When Should Public Debt Be Reduced? I M F S T A F F D I S C U S S I ON N O T E When Should Public Deb Be Reduced? Jonahan D. Osry, Aish R. Ghosh, and Raphael Espinoza June 2015 SDN/15/10 When Should Public Deb Be Reduced? Prepared by Jonahan

More information

Dynamic Contracting: An Irrelevance Result

Dynamic Contracting: An Irrelevance Result Dynamic Conracing: An Irrelevance Resul Péer Eső and Balázs Szenes Sepember 5, 2013 Absrac his paper considers a general, dynamic conracing problem wih adverse selecion and moral hazard, in which he agen

More information

BIS Working Papers. Globalisation, passthrough. policy response to exchange rates. No 450. Monetary and Economic Department

BIS Working Papers. Globalisation, passthrough. policy response to exchange rates. No 450. Monetary and Economic Department BIS Working Papers No 450 Globalisaion, passhrough and he opimal policy response o exchange raes by Michael B Devereux and James Yeman Moneary and Economic Deparmen June 014 JEL classificaion: E58, F6

More information