Upload
aminul-haque
View
214
Download
0
Embed Size (px)
Citation preview
Future Generation Computer Systems 27 (2011) 1056–1069
Contents lists available at ScienceDirect
Future Generation Computer Systems
journal homepage: www.elsevier.com/locate/fgcs
A survey of economic models in grid computingAminul Haque a,∗, Saadat M. Alhashmi a, Rajendran Parthiban b
a School of Information Technology, Monash University Sunway Campus, Jalan Lagoon Selatan, Bandar Sunway, 46150, Selangor Darul Ehsan, Malaysiab School of Engineering, Monash University Sunway Campus, Jalan Lagoon Selatan, Bandar Sunway, 46150, Selangor Darul Ehsan, Malaysia
a r t i c l e i n f o
Article history:Received 25 June 2010Received in revised form7 April 2011Accepted 11 April 2011Available online 16 April 2011
Keywords:Grid computingEconomic modelDynamic environment
a b s t r a c t
Grid computing offers the network of large scale computing resources. Economic models are effective incollaborating large scale heterogeneous grid resources that are typically owned by different organizations.Not all the models provide same benefits for users in utilizing the resources. Similarly, the profit earnedby resource providers also differs for different economic models. We survey the economic models used ingrid computing since its inception until 2010.We discuss their advantages and disadvantages and analyzetheir suitability for usage in a dynamic grid environment. To the best of our knowledge, no such surveyhas been conducted in the literature up to now.
© 2011 Elsevier B.V. All rights reserved.
1. Introduction
Investigation of some problems in science, engineering andcommerce such as protein analysis, material properties, and eco-nomic forecasting are computationally complex. By realizing theinsufficiency of a single computer, a cluster or even a supercom-puter in solving these problems, grid computing was initiated inthe mid 1990s [1]. The technology that aggregates distributedcomputer resources across the world is called grid computing. Co-ordination of distributed and heterogeneous computing resourcescreates virtual organizations that support the utilization of idleresources [2]. However, seamless collaboration is a challenge dueto the extreme heterogeneity of these resources. This hetero-geneity is due to varying architectures architecture of physicalresources (e.g. clusters, supercomputers, ordinary PCs), differentadministrative domains (e.g. country, enterprize) andmultiple op-erating systems (e.g. UNIX variants, Windows). There is also a lackof a uniform way to use these resources.
Fig. 1 shows layers and different components that constitutea typical grid. The layered grid architecture usually rests onthe fabric layer that consists of servers, clusters, monitors andall other distributed computing resources around the world.Mercury [3] system is a good example for this layer. The layerthat controls and allows secure access to the components offabric layer is called the core middleware layer. It also supports
∗ Corresponding author. Tel.: +603 5516 1907; fax: +603 5514 6129.E-mail addresses: [email protected] (A. Haque),
[email protected] (S.M. Alhashmi),[email protected] (R. Parthiban).
0167-739X/$ – see front matter© 2011 Elsevier B.V. All rights reserved.doi:10.1016/j.future.2011.04.009
Fig. 1. Layered grid architecture with examples.
trading and information updating of resources. Globus [4] is awell-knownmiddleware service, which allows resource discovery,management and security. On the other hand, Gridbus [5]middleware supports business driven technologies aimed atutility based computing. Gridbus uses economic models that aidefficient management of shared resources through maintainingthe supply and demand of distributed resources. In this paper,we focus on suitable economic models in grid computing andtheir practicality of usage in different perspectives. The upperlevel of core grid middleware is called user level middlewarethat provides API (Application Programming Interface), libraries,application development environments and resource mediator,
A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069 1057
Fig. 2. A reference Market oriented overview of a layered Grid architecture [2].
which negotiates between users and providers, and schedulesapplication tasks for execution on global resources. The SimpleAPI Grid Applications (SAGA) [6] and Triana [7] are the twoexamples of user level middleware. This middleware is used tocommunicate with the core middleware. Grid applications, thefourth layer, is typically developed using the components ofuser level middleware. This layer supports users to execute theirapplications on remote resources and collect results from themusing web portals or applications such as the Grid ApplicationToolkit (GAT) [8] and java Commodity Grid kit (CoG) [9].
The core part of the grid is called core grid middleware,since it offers all the necessary functions such as scheduling,security, data transfer, trading and communication [10]. The mainobjective of this middleware is to hide the heterogeneous natureand provide a homogeneous and flexible environment to endusers. Trading is one of the main parts that motivates resourceproviders to contribute their resources on grid computing. Inaddition, price is a key deciding factor in resource use [11].Price can further be used to maintain equilibrium betweensupply and demand, distinguish different QoS (Quality of Service)requirements and utilize idle resources. A market orientedmodeling can be used in solving distributed resourcemanagementproblems such as site autonomy problem, objective optimizationproblem and cost management problem [12]. The site autonomyproblem could occur while accessing resources that belong todifferent administrative domains. The objective optimizationproblem occurs when users want to optimize their QoS andwhen providers want to maximize their profit. Grid resourceproviders need to support seamless management of differentrequests from different users simultaneously—this known as thecostmanagement problem. Several economicmodels are proposedin the literature for driving market-oriented grid computing. Oneeconomic model is different from the others in pricing resourcesfor varying scenarios.
In this article, we investigate suitable economic models forgrid computing. The analysis covers models that have beenproposed since the inception of grid computing and discusses theirstrengths and weaknesses as perceived by researchers. Finally,
we identify that different models are suitable for dealing withdifferent grid scenarios; however we find no work consideringmultiple economic models and switching between them forvarying scenarios in a grid environment.
The remainder of this article is organized as follows: InSection 2, we explore market oriented grid computing. Section 3explains the strengths and weaknesses of different economicmodels that have been proposed and used by different gridcomputing researchers thus far. Section 4 maps out some futureresearch directions in market oriented grid based on the analysesof the models.
2. Market oriented grid computing
Standardization, usability and business models have been ac-cepted as the main success factors for next generation computingsystems [13]. However, market based computing mechanisms aredifferent from the traditional mechanisms in terms of value (i.e.QoS) delivered to a user. The value could be measured by the fol-lowing:
• flexibility in parameterization of user driven jobs,• suitability of business models for different user requirements
and strategies and• adaptation to changes in resource availability, capability and
pricing.
To realize this, market oriented computing organizations needto be more complex than the traditional systems.
Fig. 2 presents the four grid layers of Fig. 1 in terms of marketoriented modeling environment. Each layer has some additionalfunctional entities along with dependencies among them. Thearrow from A to B refers to the dependency of A on B. In amarket oriented architecture, market directory service keeps theresource information updated and helps to generate a competitivemarket price for a particular type of resource. Market basedmiddleware supports market participants to trade grid resources.It performs trading activities such as SLA (Service Level Agreement)enforcement and billing, contract and trading management. All
1058 A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069
of these interact with resources to decide which resources areallocated to which user, for what price and for how long.Economically Enhanced Resource Management (EERM) isolatesusers from its providers based on certain market relevant featuresto increase its functionality. EERM also keeps itself updated withthe resource state through monitoring services and reports to theSLA enforcement. EERM gets information from supply modelingand assists to form SLA. Supply modeling depends on demandmodeling, which provides necessary tools to specify resourceproperties. Offers are generated based on both supply and businessmodels. On the other hand, bids are generated based on users’demands and preferences (e.g. economic preferences). A resourcemediator negotiates between resource users and providers. Aresource mediator could be a resource broker or a resource agent.Market oriented grids require adaptive management capabilitiesamong different functional entities to enhance the service qualitydelivered to users as well as to optimize providers’ goals.
2.1. Inspiration for economic grids
Technology, business and policy are interdependent; withouttechnology there are no services and products to be invented,without business models no policies are needed to regulatetheir actions [14]. Buyya argues that the grid’s heterogeneityand decentralization is similar to the present standard humaneconomy [15], where market based mechanisms could be used tosuccessfully manage the environment. He further argues that thisapproach would be efficient for balancing supply and demand andit is scalable (no need for central coordinator during negotiation).Additionally, it improves the utilization of idle resources anddistinguishes different quality of services. Similar measurementsof a market based grid can be seen in [16–20].
Traditional market pricing models for managing grid resourceswould also be applicable for managing self-interested and self-regulating entities (resource owners and users) [21]. A studyundertaken by Chris and Giorgos [11] demonstrates the possiblemacroeconomic value for the introduction of grid computing andforecasts a huge amount of gain through the deployment of highperformance grid computing and web service applications. Thepaper argues that the price impact could be very important forindustrial firms which use grid computing. Using grid computing,the firms become more competitive than might otherwise beexpected. Grid technology enables the compilation of resourcesacross many budget boundaries (accessing different economicgoals). Therefore, an appropriate business model would be thekey term for fair dynamic resource collaboration. Price can be akey deciding factor in resource use as well. In a market orientedapproach, uncertainty drives a large portion on the decisions suchas what is available when and for what price [11]. According toone of the leading grid computing resource institutes, The 451Group, the application of resource trading and allocation modelsis one of the crucial success factors for establishing commercialgrids [22]. Therefore, a suitable pricing model for Internet servicesis one of the main prerequisites for successfully running theimplementation of an accounting and charging system. Shin Yeoand Buyya [23] focus on a pricing mechanism to support utilitydriven management and allocation of resources. Accordingly, theproviders should have mechanisms for generic pricing schemesto increase system utilization and protocols that help them offercompetitive services.
Due to the significance of pricing in grid computing, a robustand viable economic model is required to deal with pricingdistributed resources across multiple administrative domains.In addition, economic models help providers to treat differentusers differently based on their requirements and organizecorresponding SLAs, which collectively would construct a rigid
market oriented computing environment. Buyya proposes severaleconomic models (such as commodity market, posted price, andbartering) including both micro andmacroeconomic principles fordistributed resourcemanagement [15]. However, he only discussesa hypothetical suitability of these models for a grid environment.Not all the models proposed are suitable to deal with all differentscenarios as we will see in the following sections. In Section 3,we discuss about the models that have been studied and analyzedby different grid researchers. We address the strengths andweaknesses of these models in terms of managing heterogeneousgrid resources. Before startingwith various economicmodels in thegrid, we first introduce various criteria that can be used as probesto judge economic grids.
2.2. Criteria to judge economic grids
Economicmodels are different from one another in terms of theway (i) they are used for interaction among users and providers, (ii)they are used for pricing, and (iii) they adapt to evaluate differentrequirements. In a grid computing environment, the strengths andweaknesses of an economic model can be evaluated using severalcriteria. Some of themarementioned herewith a brief explanation.Admission control: Admission control refers to the control ofsubmitting new jobs in a grid to execute. This feature plays animportant role in maintaining market equilibrium. In a marketoriented grid, dynamic pricing can be used tomaintain equilibriumbetween supply and demand that change dynamically.Broadcasting overhead: This is also known as distribution andcommunication overhead. It is the delay incurred to disseminateinformation regarding resource availability, pricing bids etc. overthe Internet. It also depends on geographic distance of computingendpoints, Internet speed and communication protocols.Computation efficiency: This is the amount of computation timethat is consumed by a particular model while evaluating users’requests. Models that consume fewer computation cycles areconsidered to be computationally efficient.Decentralization: Decentralization in an economic based grid refersto the freedom of setting a resource price by a provider. In adistributed environment such as grid computing, decentralizationin pricing is expected to achieve large scale resource collaboration.It can also be used to evaluate global allocation efficiency.Evaluating market price: The market price or economic price ofa particular resource could be manipulated by the price offeredfor the same resource by different providers. The demand on aparticular resource also contributes to determine the market priceof that resource. A true market price is crucial in achieving acompetitive grid market.Handling large number of users: This criterion refers to the abilityto contact and evaluate a large number of requests with theirdifferent QoS requirements within a particular period of time.Typically the grid resources are utilized by global users overthe Internet. Hence, there could be many users in general. Thiscriterion could also refer to the scalability of dealing with manyusers.Job cancelation rate: This is the rate at which requested orsubmitted jobs are canceled by a particularmarket. Job cancelationcould occur due to the disagreement of prices, unavailability ofresources and failure while executing the jobs in the grid.Price stability: This criterion explains the stability of a marketprice for a particular resource for a specific amount of time.Inflation is the opposite of price stability. Inflation is the raising ofresource prices. Again price stability is crucial to ensure schedulingstability [24].
A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069 1059
Pareto optimal allocation: This is a resource allocation process, inwhich allocating of a particular resource is not supposed to affectother resources that are currently being allocated or executed [25].This is necessary in grid computing to get the jobs of currentusers done by their requested deadlines. Economic models play animportant role here in setting different SLAs for different users.Utility based negotiation: This is also called individual rationality,which refers to the payoff gained through participating innegotiation. In an agent based negotiation, individual rationalitymeans, all the agents in the system agree to participate innegotiation, since each of them is individually assured of receivinga better payoff than in the case of not participating [26]. In a gridmarket, negotiation could happen between a user and a providerindividually in order to optimize their goals or objectives.Resource allocation efficiency: The ability to allocate an appropriateamount of resources according to the needs of users is calledresource allocation efficiency [27]. It would help users to gettheir jobs executed within their job deadlines. Similarly, providerswould be benefited through provisioning their resources to users.Economic efficiency: The economic efficiency of a particulareconomic model defines how efficient the model is in utilizing idleresources as well as maximizing profit for providers. Profit for aparticular period could be presented as a subtraction between thetotal revenue gained and total expenses associated with providingservices throughout that period. In fact, economic efficiencydepends on all the criteria mentioned above. In addition, in auser’s point of view, a model can be efficient if it supports theuser’s requirements. Therefore, the economic efficiency could alsobe treated as user provider efficiency. Social welfare can be usedto determine economic efficiency. Social welfare is calculated byaggregating users’ and providers’ utility, which for a particularentity (e.g. user/provider) is defined as the difference betweenhis/her reservation-price1 and agreed-price2 for a particularservice.
3. Economicmodels and their strengths andweaknesses in gridcomputing
In this section, we investigate different economic modelsproposed by different grid computing researchers since theinitiation of the grid. We briefly explain about different economicmodels, since an extensive explanation on different models hasalready been given by Buyya et al. [15]. The main focus of thissection is to present the economic models in terms of theirstrengths andweaknesses as identified by different grid researchesat different times. At first, we present the strengths of differentmodels in Table 1.
Table 1 demonstrates the significance of economic models ingrid computing. The papers are presented in descending ordersof years so that the reader can easily perceive recent works oneconomicmodels. It can be seen from the table that the Commoditymarket and DA are the most widely proposed models in thegrid. The Commodity market model has the ability to maintainmarket equilibrium, which is crucial for any market-oriented gridenvironment. Maintaining supply and demand by regulating pricebehavior ensures the higher probability to deliver requested QoSas well as increased system performance. The main purpose ofthis model is to determine a supply and demand equilibrium/spotprice. For example, if demand for a resource exceeds its supplyat a particular state, the price of that resource increases in away such that the demand function shifts to the point closer to
1 Price limit at which a user/provider has agreed to buy/sell a particular service.2 Price at which both user and provider are satisfied to trade.
the available supply. Various techniques are used to determinethe equilibrium/spot price in the literature [33]. DA, on theother hand is suitable due to its decentralized nature and theability to handle large number of users. In grid computing, usersand providers are self-interested entities and appear with theirindividual optimization strategies. Hence, DA supports them bysorting their valuations and thus expediting the trading phasewithout any requirement of global information.
English auction is another interesting model in the grid,where auctioneer seeks to obtain the true market value of theresource that has been set for auction. Usually, users are free toincrease their bids exceeding others for the resource that theyare competing for. When no bidder is willing to increase theirbids any more, the auction ends and the auctioneer checks itsreservation price with the last highest bid and determines thewinner. This model is found to be suitable for increasing revenue,since it supports competition among users and finally selects theuser who bids the highest by using an iterative bidding policy. Thisalso helps to identify the demand of a particular resource in themarket. However, English auction, in a distributed environmentmay produce network congestion due to its high communicationdemand. English auction, by nature, is an iterativemodel andhencecauses too many messages to be exchanged during the auctionprocess [55].
By using the Bargaining model in grid computing, users andproviders can optimize their preferred interests. The model allowsits participants to negotiate their preferences and finally constructa satisfactory SLA. In grid computing, the preferences could beover budget/job-execution-cost, deadline/job-execution-time orany such criteria. However, successful negotiation also depends onpreference values. For example, if a user and a provider negotiatethe same preference (e.g. deadline & job-execution-time) value,either the negotiation will finish with minimum optimizationof the preference or it will fail. The optimization of a specificpreference for a particular user canbemeasuredbyusing theutilityfunction of the optimization criteria for that user. Researchershave already analyzed how to relax different negotiation termsduring the bargaining process, so that better optimization canbe achieved [75,78]. The Bargaining model also requires highcommunication demand due to the multi-round communicationprocess whichmay not be suitable when there is a large number ofusers.
A proportional share-based allocation is efficient for gridcomputing, since it allows sharing of resources according users’demands. The model also helps to construct a large scale sharedinfrastructure which is one of the main goals of grid computing.Because of sharing the same resources by multiple users,utilization for the resources increases and hence job cancelationrate decreases. However, failure to provide a sustainable sharingmechanism may cause a lower QoS received by users or evencancelation of jobs.
CNP allows users to choose appropriate service providersbased on their varied requirements. The model permits users tooptimize their preferences (e.g. budget, deadline) by selecting oneormore appropriate providers out ofmultiple providers in the grid.Providers are allowed to cooperate among themselves in order toensure that the service for users is as per the contract. This modelbasically focuses on the users’ side to optimize their preferencesthan for the providers. Hence, the utility for users is generallygreater than that for providers.
If we observe Table 1 as a whole, we can see that differentmodels are suitable for different scenarios. For example, theCommodity market model is suitable for market equilibrium,admission control and Pareto optimal allocation, whereas theDA model shows strengths in handling a large number of users,decentralization and time efficiency. The English auction model
1060 A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069
Table1
Streng
thsof
econ
omic
mod
elsin
grid
compu
ting.
Econ
omic
mod
elStreng
ths
Prop
osed
by:r
esea
rchfocu
s/co
ntribu
tion
Commod
itymarke
t:In
gene
ral,reso
urce
sarepr
iced
insu
chaway
sothat
equilib
rium
betw
eensu
pply
andde
man
dismaintaine
d.Th
erearetw
otype
sof
Commod
itymarke
tmod
elin
gene
ral:fla
tpricing
mod
elan
dsu
pply
andde
man
dba
sedpr
icingmod
el.T
helatter
ismorepo
pular
amon
gresearch
ers,sinc
eitha
sthecapa
bilityto
maintaineq
uilib
rium
betw
eenreso
urce
supp
lyan
dde
man
dby
chan
ging
pricebe
havior
Admission
control,co
mpu
tingeffic
ienc
y,ec
onom
iceffic
ienc
y,fle
xibilityin
evalua
tingmarke
tprice
,paretoop
timal
allocatio
n,reso
urce
allocatio
neffic
ienc
y,pr
icestab
ility
Bossen
broe
ket
al.[28
]:Minim
izerisk
asso
ciated
with
serviceoffering
/req
uestingd
ueto
pricevo
latility
byad
optin
ghe
dgestrategy
a
Saur
abhet
al.[29
]:Minim
izingco
stan
dex
ecutiontim
eus
ingmetasche
dulin
ghe
uristic
sNim
rod/G[ 30]:E
cono
mic
fram
eworkfors
ervice
driven
reso
urce
sche
dulin
gKe
vin[ 31]:C
ompa
reec
onom
icba
sedap
proa
chov
erno
nec
onom
icce
ntraliz
edap
proa
ches
Chen
get
al.[32
]:Utilizegrid
reso
urce
sba
sedon
appr
opriateserviceselection
Shin
etal.[23
]:Su
pportu
serc
entric
jobsp
ecificatio
nfors
uitableallocatio
nde
cision
sStue
reta
l.[33]:A
chieve
effectiveallocatio
nwhile
ensu
ring
pricestab
ility
andservice
fairne
ssOmer
[ 34]:E
cono
mic
basedsche
dulin
gfort
imeco
stop
timizationan
dpa
rameter
swee
pap
plications
DIRSS
G[35]:M
inim
izingjobcanc
elationrate,w
hile
ensu
ring
scalab
ility
andload
balanc
eCh
unlin
[36]:O
ptim
izeag
greg
ateutiliza
tionforg
ridus
ers,while
max
imizingreve
nue
forp
rovide
rsCh
unlin
etal.[37
]:Stud
yutility
basedallocatio
nalgo
rithm
prop
ertie
sun
derb
udge
tan
dtim
eco
nstraints
Nim
rod/G,G
ridb
us[38]:S
ched
uleco
mpu
tatio
nally
complex
andda
taintens
ive
applications
unde
reco
nomic
driven
grid
Tian
field
[39]:A
gent
basedne
gotia
tionford
istributed
reso
urce
man
agem
ent
Gridm
arke
t[40
]:Effectivetask
sche
dulin
gin
supp
lyan
dde
man
ddr
iven
grid
compu
ting
Gco
mmerce
[24]:C
ompa
reco
mmod
ityan
dau
ctionpr
otoc
ols’effectiven
essin
term
sof
marke
tcon
trol
GRA
CE[41]:D
ynam
icreso
urce
trad
ingforflexibleap
plicationsche
dulin
gNim
rod/G[42]:S
ched
ulingus
ingad
aptiv
eman
agem
ento
fcom
putatio
nalresou
rces
Kurt
[43]:A
compa
rativ
estud
ybe
twee
nco
mmod
itymarke
tand
Vick
reyau
ction
mod
elsin
case
ofpr
icestab
ility,fairn
essof
allocatio
nsan
dco
mmun
ication
requ
irem
ents
Dou
bleAu
ction(D
A):P
rovide
rsarearrang
edin
asce
ndingorde
rand
users
inde
scen
ding
orde
rinterm
sof
deman
dan
dbu
dget
resp
ectiv
ely.
Ifa
user’srequ
estm
atch
eswith
apr
ovider’soffer,thetrad
eispe
rformed
.Th
erearetw
otype
sof
DA;
Continuo
usDou
bleAu
ction(CDA)
(inwhich
userspo
sttheirr
equiremen
tsan
dbu
dgets,an
dservicepr
oviderspo
sttheiro
ffers
atan
ytim
edu
ring
thetrad
ingpe
riod
)and
period
icDA
(whe
reau
ctionco
ntinue
sfora
spec
ifictim
epe
riod
asde
fined
bythe
auctione
er).Th
eform
ertype
isdiscus
sedmos
tlyin
thegrid
literatur
e
Marke
tcom
petit
ion,
reso
urce
allocatio
neffic
ienc
y,broa
dcastin
gov
erhe
ad,
compu
tingeffic
ienc
y,ha
ndlin
glarge
numbe
rofu
sers,p
rice
stab
ility,
dece
ntraliz
ation,
econ
omic
effic
ienc
y
Izak
ianet
al.[44
]:Max
imizetask
completionrate,u
tilizationof
reso
urce
san
dpr
ofit
forp
rovide
rsLi
etal.[45
]:Su
pportc
ombina
torial
bids
andex
hibitinc
entiv
ech
aracteristicsforb
oth
usersan
dpr
oviders
Wan
g[46]:M
otivateus
ersan
dpr
ovidersthroug
hsu
pportin
gindividu
alratio
nality
Lyna
reta
l.[47]:S
tudy
variationof
timean
den
ergy
cons
umptionby
applying
diffe
rent
auctionpr
otoc
ols
Suri[48]:M
aintainpr
icestab
ility
usingkn
owledg
eba
sedpo
licy
Valken
hoef
etal.[49
]:Co
mpa
reTC
DAb
with
trad
ition
alCD
Ain
term
sof
exec
ution
unce
rtainty
Wieczorek
etal.[50
]:Stud
yworkflow
beha
vior
tosu
pportfastera
ndch
eape
rex
ecution
Streitb
erge
reta
l.[51]:C
ompa
rece
ntraliz
edan
dde
centraliz
edserviceallocatio
nin
term
sof
timean
dutility
func
tion
A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069 1061
Table1(con
tinue
d)
Econ
omic
mod
elStreng
ths
Prop
osed
by:r
esea
rchfocu
s/co
ntribu
tion
Pour
ebrahimie
tal.[52]:D
ecisionmak
ingag
ents
adap
ttody
namic
netw
ork
environm
enta
ndpr
icing
Tan[ 53]:Inv
estig
atemarke
tprice
throug
hpr
ovidingabidd
ingad
justmen
tstrateg
yW
enget
al.[54
]:Re
sour
cetype
basedmod
elingto
supp
ortd
ynam
icad
justmen
tof
pricing
Marco
s[ 55]:C
ompa
reco
mmun
icationov
erhe
adan
dpr
ofitford
ifferen
tauc
tion
mod
els
Pour
ebrahimie
tal.[ 56]:S
tudy
pricingfunc
tionin
balanc
edan
dun
balanc
edne
tworks
with
selfinterested
agen
tsSX
[ 57]:S
uppo
rtorga
niza
tions
tofede
rate
storag
eservices
amon
gthem
andlease
them
glob
ally
Eyman
net
al.[58
]:DesignAL
Ncforb
othce
ntraliz
edan
dde
centraliz
edorga
niza
tions
Kant
[59]:A
compa
rativ
eap
proa
chof
diffe
rent
DApr
otoc
olsto
max
imizereso
urce
utiliza
tion
Nim
rod/G,G
ridb
us[38]:S
ched
uleco
mpu
tatio
nally
complex
andda
taintens
ive
applications
unde
reco
nomic
driven
grid
BrickW
orld
[25]:M
ultip
lesing
leite
mau
ctionto
avoidco
mplex
itywith
combina
torial
bids
Gridm
arke
t[40
]:Stud
ypr
icingalgo
rithmsforu
sers
andpr
oviderssepa
rately
tomea
sure
theintegrity
ofrequ
ests
andoffers
Grosu
[60]:S
tudy
econ
omic
effic
ienc
yan
dsystem
performan
ceof
threeau
ction
mod
els
Gom
oluc
h[61]:S
tudy
system
load
,heterog
eneity
andco
mmun
icationde
laywith
threediffe
rent
mod
els
Heet
al.[62
]:Man
ipulatean
dad
apttody
namic
marke
tprice
usingfuzzylogic
Englishau
ction:
Acco
rdingto
thistype
ofau
ction,
usersarefree
toincrea
setheirb
idsov
ertaking
othe
rs.W
henno
newou
ldlik
eto
increa
sethepr
icean
ymore,
theau
ctionen
ds.T
heau
ctione
erde
clares
thehigh
est
bidd
eras
winne
r.Bids
canbe
prop
osed
fora
sing
leite
m(singleattribute)
orform
ultip
leite
ms(m
ultiattributes)
QoS
,eco
nomic
effic
ienc
y,reve
nue,
reso
urce
allocatio
neffic
ienc
y
[63]:D
evelop
reso
urce
map
ping
algo
rithmsus
ingite
rativ
eco
mbina
torial
auctionmec
hanism
MAC
E[64]:D
esignan
auctionba
sedco
nstruc
tiveec
onom
icmod
elto
assist
grid
users
inex
pressing
theirt
ruede
man
dsBe
cket
al.[17
]:Stud
yec
onom
iceffic
ienc
ywhile
prov
idingsu
itableallocatio
nan
dlearning
mod
els
GEM
SS[65]:Implem
enta
simulationtoolkitc
onside
ring
therising
need
sfrom
med
ical
services
Attana
sioet
al.[66
]:Dev
elop
auctionmec
hanism
s,while
ensu
ring
minim
alco
mmun
icationov
erhe
adwith
effic
ient
reso
urce
usag
eEG
G[67]:S
implify
jobsche
dulin
gthroug
han
dec
onom
icplatform
inde
centraliz
edgrids
BrickW
orld
[25]:A
uctio
nwith
man
yite
msto
avoidco
mplex
ityin
combina
torial
bids
Tian
field
[39]:S
tudy
agen
ttec
hnolog
yfora
daptive,
runtim
eeffic
ienc
yan
dau
tono
mou
sgrid
Bella
gio[68]:A
nalyze
scalab
ility,e
fficien
cyan
dlong
term
beha
vior
forr
esou
rces
allocatedin
fede
ratedgrid
Bargaining
:Inthismod
el,u
sers
liketo
getlow
eracce
sspr
icean
dhigh
erus
agedu
ratio
n.Th
epr
oviderslik
eto
getm
orepr
ofitthroug
hba
rgaining
.Th
eus
ersmight
startw
ithave
rylow
pricean
dpr
oviderswith
ahigh
erpr
ice.
Bargaining
may
continue
over
multip
leattributes
(e.g.p
rice
,de
adlin
e/job-ex
ecution-
time)
Utility
basedne
gotia
tion
Subrataet
al.[69
]:Dev
elop
semi-static
sche
dulin
galgo
rithm
tomax
imizeutility
for
prov
iders
(con
tinue
don
next
page)
1062 A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069
Table1(con
tinue
d)
Econ
omic
mod
elStreng
ths
Prop
osed
by:r
esea
rchfocu
s/co
ntribu
tion
Zhao
[70]:M
aintainmarke
tequ
ilibrium
andmax
imizepr
ofitthroug
hself-
adap
tive
autono
mou
sne
gotia
tion
Qua
n[ 71]:D
eterminemarke
tprice
basedon
dead
line,
urge
ncyof
workflow
man
agem
enta
ndgrid
state
Assu
ncao
[72]:O
ptim
izereso
urce
utiliza
tionan
dload
balanc
eacross
fede
ratedgrids
ABRM
AS[ 73]:A
nalyze
agen
tbased
mec
hanism
ofreso
urce
discov
eryforins
uffic
ient
budg
eted
users
GridSim
[74]:Implem
enta
simulationen
vironm
ents
uitableforu
tility
basedgrid
compu
ting
Sim
[75]:D
etermineap
prop
riateam
ount
ofrelaxa
tionin
nego
tiatio
ncrite
riato
max
imizeutility
andsu
ccessrate
Li[76]:A
nalyze
timean
dlearning
basedne
gotia
tionstrategies
fora
daptingwith
dyna
mic
grid
Gho
shet
al.[77
]:Harne
ssco
mpu
tingpo
wer
inmob
ilede
vice
sthroug
han
effic
ient
pricingstrategy
toallocate
jobs
onthem
Sim
[78]:S
tudy
bargaining
mod
elsby
cons
ideringgrid
dyna
micsan
dap
prop
riate
relaxa
tionof
bargaining
term
s
Prop
ortio
nalsha
reba
sedau
ctiond
:Ec
onom
iceffic
ienc
y,scalab
ility
Tyco
on[79]:A
llocate
hostseffic
iently
inaclus
ter.Dev
elop
anag
entb
ased
appr
oach
toallocate
reso
urce
sXa
vier
etal.[80
]:Man
agereso
urce
load
ondistribu
tedlargescaleinfrastruc
ture
byco
ntrolling
reso
urce
prices
across
agrid
netw
ork
Prop
ortio
nalresou
rceshare:
Inthismec
hanism
,the
percen
tage
ofthe
reso
urce
shareallocatio
nto
theus
er’sap
plicationispr
oportio
naltothe
bidva
luein
compa
riso
nto
theothe
rusers’b
ids
Less
jobcanc
elationrate
OSE
P[81]:A
llocate
reso
urce
sfairly
throug
how
ners
hare
enforcem
entp
olicyan
ddistribu
tedow
nershipco
ncep
tGom
oluc
h[61]:S
tudy
andco
mpa
reDAan
dpr
oportio
nalresou
rcesh
arein
term
sof
system
load
,heterog
eneity
andco
mmun
icationde
lay
Li[82]:A
pply
agen
ttec
hnolog
yto
max
imizejobacco
mplishm
entr
atewhile
minim
izingtheco
staccrue
dLibra[83]:S
hare
dead
linean
dbu
dget
strategically
byco
nsideringus
erutility
rather
than
system
performan
ceFirstp
ricesealed
bidau
ctione
:Re
sour
ceallocatio
neffic
ienc
y,glob
alallocatio
neffic
ienc
yMirag
e[84]:D
eploytestbe
dreso
urce
sto
compu
tingus
ersin
anag
greg
ated
man
ner
throug
hco
mbina
torial
auction
A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069 1063
Table1(con
tinue
d)
Econ
omic
mod
elStreng
ths
Prop
osed
by:r
esea
rchfocu
s/co
ntribu
tion
Contract
netp
rotocol(CN
P):A
ccording
tothismod
el,a
user
iscalle
da
man
ager
andapr
ovider
iscalle
daco
ntractor.H
ere,
aman
ager
declares
his/he
rreq
uiremen
tsan
dinvitesbids
from
availableco
ntractors.
Interested
contractorsev
alua
tethede
man
dsan
dresp
ondby
subm
itting
theirb
ids.Th
eman
ager
evalua
testhebids
andselectsaco
ntractor
topr
ocee
d
Utility
basedne
gotia
tion,
scalab
ility,
reso
urce
coop
eration,
meta-sche
dulin
g
Chao
etal.[85
]:Group
grid
node
sin
term
sof
theirr
espe
ctivede
siresto
optim
ize
system
performan
ceCa
ramia
[86]:O
ptim
izesystem
performan
cethroug
hne
gotia
tingdistribu
ted
sche
dulers
Gos
wam
i[87
]:Max
imizesu
ccessrate
while
minim
izingtim
ean
dco
stco
nstraintsat
diffe
rent
jobarriva
lperiods
Stefan
o[88]:O
ptim
izeQoS
byad
optin
gCD
Nfco
ncep
tingrid
Ranjan
etal.[89
]:Optim
izeQoS
andreso
urce
allocatio
nde
cision
sby
SLAba
sedsu
per
sche
dulin
gin
fede
ratedgrids
Dom
iniaket
al.[90
]:Im
plem
entC
ICgto
facilitateform
ingteam
sof
diffe
rent
prov
isioning
andsp
ecializ
ation
Paur
obally
[91]:D
evelop
multi-
agen
tneg
otiatio
ntech
niqu
esto
facilitatebu
ildingan
adap
tivean
dau
tono
mou
sgrid
Oue
lhad
jeta
l.[92]:D
esignSL
Aba
sedne
gotia
tionto
deal
with
unce
rtaintiesin
reso
urce
coop
eration,
system
flexibilityan
dscalab
ility
Juha
sz[93]:S
tudy
system
performan
cein
term
sof
system
size
,age
ntload
and
dead
line
Poojaet
al.[94
]:Prop
osemulti-
agen
tbased
hierarch
ical
bidd
ingmec
hanism
tometa-sche
dule
grid
services.S
uppo
rtre-n
egotiatio
nam
ongag
ents
subjec
tto
unce
rtaintyof
jobex
ecution
aMak
eco
ntractsto
obtain
therigh
tsof
buying
/sellin
gapa
rticular
servicewith
inasp
ecificpe
riod
andat
asp
ecificpr
ice.
bTrus
tbased
Continuo
usDou
bleAu
ction:
supp
orts
agen
tsto
committo
trad
esthey
trus
t.cAp
plicationLa
yerN
etwork:
hide
sthehe
teroge
neity
ofaservicene
tworkfrom
users’view
.dTh
isislik
etheEn
glishau
ction,
exce
pt,a
fter
theau
ctionpr
ocess,reso
urce
saresh
ared
amon
gthepa
rticipan
tsacco
rdingto
theirb
ids.
eAnu
mbe
rofu
sers
subm
ittheirb
idson
lyon
ceto
geta
service,
with
outk
nowingothe
rbids.Th
ehigh
estb
idde
rwinstheserviceat
thepr
icehe
/she
bids
.fCo
nten
tDistributionNetwork:
duplicates
web
reso
urce
s(owne
dby
thesameorga
niza
tion)
from
anorigin
server
todiffe
rent
replicaservers.
gClient
Inform
ationCe
nter:s
toresinform
ationso
that
aus
erag
entc
aninteract
andne
gotia
teon
pre-ex
ecutionen
tities(e.g.p
rice
,QoS
)orjoinan
agen
ttea
m.
1064 A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069
is suitable for optimizing QoS related to jobs and maximizingrevenue for providers. Additionally, it is suitable for efficientresource allocation. The Bargaining model and CNP support utilitybased negotiation. CNP can further help in cooperating distributedresources and maintaining scalability. The Proportional sharebased auction model is suitable for economical efficiency andrevenue,whereas the Proportional resource sharemodel decreasesthe job cancelation rate. The First price sealed bid auction modelprovides a globally efficient resource allocation.
Apart from this, Neumann et al. [2] mainly identify twomodes of applications: batch3 and interactive4. They furtherdistinguish grid markets in terms of application dependencywhich considers complex services and application independencywhich considers only physical resources. They propose differenteconomic models for different mode of applications. For example,for Batchmodeunder application-dependentmarket, they proposeeither multi-attribute combinatorial auction or proportional sharebased auction and for the same mode under an application-independentmarket, theypropose a bargainingprotocol. However,their proposed market mechanisms for different classes are just ahypothesis and not based on any experimental proof. Applicabilityof a market mechanism in a large scale distributed environmentrequires extensive study on the mechanism with real parameters.In this paper, we focus only on the models that have already beenstudied extensively by the researchers using real parameters.
In the literature, we also find that onemodel is being comparedto another model using various criteria. Table 2 describes this indetail: the first column presents different economic models, thesecond column shows themodels that are being compared and thelast column shows the different features used for comparison:
Table 2 explains different economic models in the grid andcompares them with one another in terms of various criteriasuch as market equilibrium, handling a large number of users anduser and provider efficiency. One of the widely proposed models(according to Table 1), the Commoditymarketmodel is better thanthe English auction model in terms of having less complexity inselecting a market to participate, handling large number of usersand maintaining market equilibrium. The Commodity marketmodel is more efficient in managing time and handling largenumber of users than the Dutch auction model. The Commoditymarket model is also more suitable than the Flat pricing modelfor resource allocation efficiency, time efficiency and scalability.Another widely proposed model, the DA, is better than the Dutch,English, First price sealed bid and Vickrey auction models in termsof maintaining market equilibrium, broadcasting overhead andachieving user and provider efficiency. The DA has better marketequilibrium, user provider efficiency and price stability comparedto the Proportional sharemodel. The DA is also better than anotherpopular model, the Commodity market, in terms of time efficiencyanddecentralization. The English auction (multi attribute)model isbetter than the Flat pricing andUnit (fixed) pricingmodels in termsof QoS and economic efficiency. Themulti attribute English auctionmodel is better than the single attribute English auction model interms of QoS optimization and consideration of combinatorial bids.The English auction model (multi attribute) is better than the Firstprice sealed bid auction, Vickrey auction model and DA modelsin terms of QoS, revenue and economic efficiency. Even thoughthe Commodity market model is one of the widely proposedmodels, it is less efficient to evaluate the true market value of aresource compared to the English auction (multi-attribute) model.
3 Planned execution time and expected termination time for this type ofapplications are possibly known in advance.4 Planned execution time and expected termination time for this type of
applications are usually known in advance.
The Proportional share based auction model is better than theEnglish and the Vickrey auction models in terms of scalability andresource allocation efficiency.
By analyzing Tables 1 and 2, it can be seen that even thoughthe economic models play an important role in grid computing,a particular model is not suitable for all the scenarios in a gridenvironment. In addition, due to the dynamic nature of the grid,the application of a single model might not be able to harnessthe full potential from the grid. However, in the literature, thereis no such mechanism of combining two or more economicmodels to utilize the strengths of multiple economic models indifferent scenarios. For example, there is no mechanism thatchooses the Commodity market model for market equilibrium andswitches to the auction model to give more profit to providers.However, managing more than one model in a highly dynamicand heterogeneous environment poses another challenge. Thefollowing section summarizes future directions that could be takenbased on our findings that would ultimately help in building arobust and viable market-based grid environment.
4. Discussion and future research directions
Since the initiation of grid computing, a number of economicmodels have been proposed for grid computing. However, notall the models are suitable for all scenarios. Through numerousresearches, experiments and simulations, only a few of them havebeen shown to be effective in grid environment. In addition, onemodel is different from another due to its distinct features andobjectives of usage. Fig. 3 summarizes different economic modelsthat have been proposed over the years for usage in the grid.
The adoption of economic based approaches started mainly atthe beginning of this decade. Hence, Fig. 3 shows papers from2000 onwards. Fig. 3 indicates that the significance of economicmodels is increasing every year. Adoption of the Commoditymarket model started in 2000 and continues until 2010 with agap in 2002 and 2003. Many of the papers on the Commoditymarket model have been published in 2007, which is quite recent.The Commodity market model has the potential of maintainingequilibrium between supply and demand, and it is economicallyefficient. This provides an incentive to resource providers tocontribute their resources in grid. The Double Auction model (DA)is another widely proposed model since 2003. In 2009, DA isthe most frequently proposed model. The DA has become morepopular especially due to its ability in handling a large numberof participants, while producing less communication overhead.However, the DA is not that economically efficient compared tothe Commodity market model or the English auction model. TheEnglish auction model achieves popularity in the grid due to itsefficient resource allocation and economic efficiency. Hence, theEnglish auctionmodel has been continuously proposed since 2004.However, the English auction model is not suitable for handlinga large number of users and is not decentralized. The Bargainingmodel has been proposed since 2005 and became popular, sinceit supports negotiation among grid participants, which assiststo form utility based computing. In spite of the utility basednegotiation, the Bargaining model is not so economically efficientand produces high communication overhead. The next threemodels, proportional share based auction, proportional resourceshare and first price sealed bid auction could not achieve thatmuchpopularity in the grid. Only a few papers have proposed thesemodels across the years. Among these three models, proportionalresource share is the most discussed, since it supports fairnessamong the resources. Finally, Contract Net Protocol (CNP) has somepotential to meet the vision of large scale resource collaboration,since it supports cooperation among different grid organizationsto optimize resource QoS. However, cooperation among the
A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069 1065
Table 2A comparative view among different economic models in grid computing.
Economic model Compared model Features
Commodity market [24,33,43]
English auction [68]
Commodity market model is:• less complex for selecting a market toparticipate• more efficient in dividing the budget, if userswant to explore different markets• more time efficient• more efficient in handling large number ofusers• more suitable for price stability• more suitable for retaining market equilibrium• more suitable for increasing user providerefficiency• more scalableEnglish auction model is:• able to evaluate market price
Dutch auction: The auctioneer begins with a high price for a particularservice, which is lowered until (a) some users are willing to accept theauctioneer’s price or (b) the provider’s minimum demand is met
Commodity market model is:• more time efficient• more efficient in handling large number ofusers
Vickrey auction, [43] Proportional resource share Vickrey auction: This isvery similar to the first price sealed bid auction model, except thehighest bidder wins at the price of the second highest bidder
Commodity market model is:• more suitable for price stability• more suitable for retaining market Equilibrium• increases user provider efficiencyVickrey auction model is:• more suitable for handling hugecommunication demand
Flat pricing (deadline and budget based)
Commodity market model is:• more efficient for resource allocation• more time efficient• more scalable
Double auction [55,60,53]
Dutch auction, English auction, First price sealed bid auction, Vickreyauction
Double auction model:• is more suitable for retaining marketEquilibrium• is more efficient for resource allocation• produces less broadcasting overhead• is more time efficient• is more suitable for handling large number ofusers• is more suitable for price stability• is more suitable for increasing user andprovider efficiency• is more efficient for global resource allocationEnglish auction model is:• more suitable to optimize QoS• more efficient for maximizing revenue forproviders• better for economical efficiency
Proportional resource share
Double auction model is:• more suitable for price stability• more suitable for retaining market Equilibrium• more suitable for increasing user and providerefficiency
Commodity marketDouble auction model is:• more time efficient• more decentralized
English auction (multi attribute) [63,64]
Flat (fixed) pricing, Unit pricingEnglish auction model (multi attribute) is:• more suitable to optimize QoS• better for economic efficiency
English auctionEnglish auction model (multi attribute) is:• more suitable for considering combinatorialbids• more suitable to optimize QoS
First price sealed bid auction, Vickrey auction
English auction model (multi attribute) is:• more suitable to optimize QoS• more efficient for maximizing revenue forproviders• better for economic efficiency
Proportional share based auction English auction, Vickrey auctionProportional share based auction model is:
(continued on next page)
1066 A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069
Table 2 (continued)
Economic model Compared model Features
• more scalable• more suitable for efficient resource allocation
Proportional resource share [83] PBS (Portable Batch System), FIFO (First in First out) Proportional resource share model has:• lower job cancelation rate
Fig. 3. Adoption of economic models as per year.
organizations would be a complex undertaking in the grid dueto their distinct administrative rules and policies. In addition,the CNP provides incentives to users through optimizing theirutility entities such as time, QoS, and budget, but does not providesufficient motivations for providers to achieve their goals. In 2008,most papers proposed the CNP model.
All economic models can be broadly categorized into twoapproaches: Commodity and Auction market. In the Commoditymarket, where price works as the main actor to regulate marketbehavior, it can be adapted to satisfy grid users [95,33]. However,price volatility in such a market is also anticipated as a harmfulcatalyst, since it might degrade the users’ QoS [28]. To avoidprice volatility, different hedging strategies are proposed, thoughconstructing a hedging portfolio by the contract issuing service forindividuals based on currentmarket conditions obviously deservesmeaningful consideration due to the higher level of uncertaintyinvolved in the system. On the other hand, different auctions aresuitable for the distributed pricing environment. Auction modelscannot always guarantee market efficiency and thus it is difficultto maintain the consistency in supply and demand. It has alsobeen identified that an individual auction model is not suitableto construct a precise and complete solution for a large scaledistributed system [95].
Though the demand for economicmodels increases over time ingrid computing, an individualmodel cannot provide all the benefitsin different scenarios. Hence, we propose the following possibleresearch directions for market based grid computing.
First, we propose a comprehensive framework that couplessuitable economic models proposed for the grid and is ableto switch from one model to another to cope with differentscenarios. For example, if a grid provider runs a Commoditymarket model for market equilibrium, it can switch to a DA, oncethe number of users is increased, since the DA is suitable forthis scenario. The framework could switch back to the previousmodel, if the provider wants to receive an equilibrium betweensupply and demand or notices that his/her available resources arediminishing. If a provider wants to contribute his/her resources in
a federated grid, he/she can switch to CNP or such an equivalentapplication environment that supports cooperation of resourcesamong different providers. Through a switching mechanism, gridproviders would get sufficient motivations to contribute theirresources on the grid, since it would help to cope with theheterogeneity and the dynamic nature of the grid and helpto make a considerable amount of profit through utilizing idleresources. However, due to the extreme dynamicity involvedin grid computing and considerable level of autonomy requiredto form some special features (e.g. determine spot price ornegotiation over entities) associated with the economic models,integrating some intelligent behavior in the environment couldbe beneficial. Switching between models also need to be happenautomatically based on current scenario, preferences (e.g. budgetoptimization, social welfare), which are usually predefined by gridusers or providers. Switching must also be conducted without anyconsiderable delay.
Second, we are proposing the agent technology to be integratedin grid computing. Agents are able to sense a particular scenarioand switch to a particular model that suits the scenario bestwithout any human intervention, since agents are distributed innature and autonomous and intelligent in behavior. The efficiencyof agent technology in grid computing has already beendiscovered,since both of them aim to achieve a large scale open distributedsystem [96]. In addition, agents canmake crucial decisions (e.g. re-track routing) during unpredicted failures (e.g. network failure)without any considerable delay in order to optimize systemutilization.
Finally, our proposed agent based switching model might notonly be useful for grid computing, but also in other conventionaldistributed systems that aim to collaborate computer and internetservices over the Internet (such as cluster computing, utilitycomputing and cloud computing). In general, all of these othertechnologies have evolved under the support of grid computinginfrastructure [97]. Due to the improvement of technologies andinexpensiveness of resources, it has been easier for the researchcommunity to define them distinctively. Cloud computing is the
A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069 1067
most recent among themand can be differentiated fromothers dueto its focus on additional scalability, dynamic configuration andvirtualized services with a greater pace. According to Foster et al.,cloud computing is driven by economies of scale, that is, achievinga cost advantage through large scale resource collaboration [97].Economicmodels can also support tradeoff decisions (e.g. betweentime and cost) and resolve risks regarding investing huge amountsof money for buying and installing real servers [98]. From a user’spoint of view, this cost reduction would help them to get low costproviders. As a result, the Provider community would be morecompetitive than might otherwise be expected. Hence, economicmodels need to be configured in such a way they can be adaptiveto the highly dynamic nature of cloud computing. Hence, ourproposed switching framework might be beneficial for the cloudcommunity as well.
5. Conclusions
Grid computing uses large scale resource collaboration. Build-ing models to fulfill this collaboration requires considerable moti-vation from resource providers. Economicmodels are proposed forthis purpose. Different economic models have been proposed overtime for the grid.We conducted an extensive survey on thesemod-els and presented their strengths and weaknesses in different sce-narios as identified by grid researchers. We observed that differentmodels are suitable for different scenarios and provided a compar-ison of their performance under these scenarios. We indicated thepossibility of switching between models to maximize benefits forproviders and /or users and highlighted the use of agents for dy-namic switching. Finally, we discussed the feasibility of using thisswitching framework in paradigms such as cloud computing.
References
[1] D.D. Roure, M.A. Baker, N.R. Jennings, The evolution of the grid, in: F. Berman,G. Fox, T. Hey (Eds.), Grid Computing: Making the Global Infrastructure aReality, John Wiley & Sons Ltd., 2003, pp. 65–100.
[2] D. Neumann, J. Stößer, C. Weinhardt, J. Nimis, A framework for commercialgrids—economic and technical challenges, Journal of Grid Computing 6 (2008)325–347.
[3] P. Kacsuk, G. Dozsa, J. Kovacs, R. Lovas, N. Podhorszki, Z. Balaton, G. Gombas,K. Peter, P-GRADE: a grid programming environment, 2008.
[4] F. Ian, Globus toolkit version 4: software for service-oriented systems, Journalof Computer Science and Technology 21 (2006) 513–520.
[5] B. Rajkumar, V. Srikumar, R. Rajiv, 26 The GRIDBUS middleware for market-oriented computing, 2009.
[6] G. Tom, J. Shantenu, K. Hartmut, K. Thilo, K. Pascal, L. Gregor Von, L. Craig, M.Andre, R. Hrabri, S. John, SAGA: a simple API for grid applications. High-levelapplication programming on the grid, 2008.
[7] T. Ian, W. Ian, S. Matthew, M. Shalil, Distributed computing with triana onthe grid, Concurrency and Computation: Practice and Experience 17 (2005)1197–1214.
[8] G. Allen, K. Davis, T. Goodale, A. Hutanu, H. Kaiser, T. Kielmann, A. Merzky, R.van Nieuwpoort, A. Reinefeld, F. Schintke, T. Schott, E. Seidel, B. Ullmer, Thegrid application toolkit: toward generic and easy application programminginterfaces for the grid, Proceedings of the IEEE 93 (2005) 534–550.
[9] V. Snigdha, G. Jarek, L. Gregor Von, P. Manish, A java commodity grid kit, 2001.[10] C. Alexandru, M. Altmann, A pricing information service for grid computing,
in: Proceedings of the 5th International Workshop on Middleware for GridComputing: Held at the ACM/IFIP/USENIX 8th International MiddlewareConference, ACM, Newport Beach, California, 2007, J..
[11] K. Chris, C. Giorgos, Grid resource commercialization: economic engineeringand delivery scenarios, IBM Zurich Research Laboratory (2003).
[12] E. Carsten, Y. Ramin, Applying economic scheduling methods to gridenvironments, in: Grid Resource Management: State of the Art and FutureTrends, Kluwer Academic Publishers, 2004, pp. 491–506.
[13] F. Neubauer, A. Hoheisel, J. Geiler, Workflow-based grid applications, FutureGeneration Computer Systems 22 (2006) 6–15.
[14] T. Eymann, et al., A framework for trust and reputation in grid environments,Journal of Grid Computing 6 (2008) 225–237.
[15] R. Buyya, et al., Economic models for resource management and schedulingin grid computing, Concurrency and Computation: Practice and Experience 14(2002) 1507–1542.
[16] H.H. Huang, et al., You can’t always get what you want: achieving differenti-ated service levels with pricing agents in a storage grid, in: Proceedings of theIEEE/WIC/ACM International Conference on Web Intelligence, IEEE ComputerSociety, 2007.
[17] B. Roman, S. Michael, H. Oliver, Grid economics in departmentalizedenterprises, Journal of Grid Computing 6 (2008) 277–290.
[18] T. Sandholm, K. Lai, Market-based resource allocation using price predictionin a high performance computing grid for scientific applications, in: 15th IEEEInternational Symposium on High Performance Computing, 2006.
[19] R. Buyya, Economic-based distributed resource management and schedulingfor grid computing, Ph.D. Thesis, Monash University, Melbourne, 2002.
[20] K. Czajkowski, I. Foster, N. Karonis, C. Kesselman, S.Martin,W. Smith, S. Tuecke,A Resource Management Architecture for Metacomputing Systems, Springer,Berlin, Germany, 1998, 62–82.
[21] R.B. Cohen, E. Feser, Grid computing, projected impact in north carolina’seconomy and broadband use through 2010, Rural Internet Access Authority,2003.
[22] W. Fellows, et al. Grid computing–the state of themarket the 451 Group, 2007.[23] C. Shin Yeo, R. Buyya, Pricing for utility-driven resource management and
allocation in clusters, International Journal of High Performance ComputingApplications 21 (2007) 405–418.
[24] W. Richard, S.P. James, B. John, B. Todd, G-commerce: market formulationscontrolling resource allocation on the computational grid, in: Proceedingsof the 15th International Parallel & Distributed Processing Symposium, IEEEComputer Society, 2001.
[25] G. Peter, Distributed combinatorial resource scheduling, in: In Proceedings ofthe 1st International Workshop on Smart Grid Technologies, SGT-2005, 2005.
[26] K.D. Rajdeep, R.J. Nicholas, C.P. David, Computational-mechanism design: acall to arms, IEEE Intelligent Systems 18 (2003) 40–47.
[27] O. Ardaiz, P. Artigas Vidal, T. Eymann, F. Freitag, L. NavarroMoldes,M. Reinicke,The Catallaxy Approach for Decentralized Economic-Based Allocation in GridResource and Service Markets, Springer, 2006.
[28] A. Bossenbroek, A. Tirado-Ramos, P.M.A. Sloot, Grid resource allocation bymeans of option contracts, IEEE Systems Journal 3 (2009) 49–64.
[29] S.K. Garg, R. Buyya, H.J. Siegel, Time and cost trade-off management forscheduling parallel applications on utility grids, Future Generation ComputerSystems 26 (2009) 1344–1355.
[30] B. Rajkumar, A. David, The Nimrod/G grid resource broker for economic-basedscheduling, in: Rajkumar Buyya, Kris Bubendorfer (Eds.), Market-OrientedGrid and Utility Computing, in: Wiley Series on Parallel and DistributedComputing, 2009.
[31] K.L. Mills, C. Dabrowski, Can economics-based resource allocation proveeffective in a computation marketplace? Journal of Grid Computing 6 (2008)291–311.
[32] W.-C. Ni, L.-C. Liu, C. Wu, Services selection based on commodity-market ingrid workflow, Journal of Computer Applications 27 (2007) 2973–2975.
[33] G. Stuer, K. Vanmechelen, J. Broeckhove, A commodity market algorithm forpricing substitutable grid resources, Future Generation Computer Systems 23(2007) 688–701.
[34] O.O. Sonmez, A. Gursoy, A novel economic-based scheduling heuristic forcomputational grids, International Journal of High Performance ComputingApplications 21 (2007) 21–29.
[35] L. Li, Y. Shoubao, DIRSS-G: an intelligent resource scheduling system for gridenvironment based on dynamic pricing, International Journal of InformationTechnology 12 (2006).
[36] C. Li, L. Li,Multi economic agent interaction for optimizing the aggregate utilityof grid users in computational grid, Applied Intelligence 25 (2006) 147–158.
[37] C. Li, L. Li, Z. Lu, Utility driven dynamic resource allocation using competitivemarkets in computational grid, Advances in Engineering Software 36 (2005)425–434.
[38] R. Buyya, D. Abramson, S. Venugopal, The grid economy, Proceedings of theIEEE 93 (2005) 698–714.
[39] H. Tianfield, Towards agent based grid resource management, in: IEEEInternational Symposium on Cluster Computing and the Grid, CCGrid 2005,vol. 591, 2005, pp. 590–597.
[40] C. Ming, et al., Gridmarket: a practical, efficient market balancing resourcefor grid and P2P computing, in: Lecture Notes in Computer Science, Springer,2004.
[41] R. Buyya, et al. An economy driven resource management architecturefor global computational power grids, in: In proceedings of internationalconference on parallel and distributed processing techniques and applications,Las Vegas, 2000.
[42] B. Rajkumar, et al., Nimrod/G: an architecture for a resource managementand scheduling system in a global computational grid, in: 4th InternationalConference on High Performance Computing, IEEE Computer Society, USA,Asia-Pacific Region, Beijing, China, 2000.
[43] V. Kurt, B. Jan, A comparative analysis of single-unit vickrey auctions andcommodity markets for realizing grid economies with dynamic pricing,in: Proceedings of the 4th International Conference on Grid Economics andBusiness Models, Springer-Verlag, Rennes, France, 2007.
[44] H. Izakian, A. Abraham, B.T. Ladani, An auction method for resource allocationin computational grids, Future Generation Computer Systems 26 (2009)228–235.
[45] L. Li, Y.-a. Liu, K.-m. Liu, X.-l. Ma, M. Yang, Pricing in combinatorial doubleauction-based grid allocationmodel, The Journal of China Universities of Postsand Telecommunications 16 (2009) 59–65.
[46] W. Yajuan, W. Xianjia, A novel double auction mechanism based resourceallocation in the grid, in: International Conference on Management of e-Commerce and e-Government, ICMECG’09, 2009, pp. 420–423.
[47] M.L. Timothy, D.H. Ric, S. Simon, Auction resource allocation mechanisms ingrids of heterogeneous computers, WSEAS Transactions on Mathematics 8(2009) 1671–1680.
1068 A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069
[48] P.K. Suri, M. Singh, Auction oriented resource allocation in grid usingknowledge based pricing policy, International Journal of Computer Scienceand Network Security 9 (2009).
[49] G. van Valkenhoef, S.D. Ramchurn, P. Vytelingum, N.R. Jennings, R. Verbrugge,Continuous double auctions with execution uncertainty, in: Workshop onTrading Agent Design and Analysis (TADA-09), Passadena, California, USA,2009.
[50] W. Marek, P. Stefan, P. Radu, F. Thomas, Applying double auctions forscheduling of workflows on the grid, in: Proceedings of the 2008 ACM/IEEEConference on Supercomputing, IEEE Press, Austin, Texas, 2008.
[51] S. Werner, H. Sebastian, E. Torsten, S. Bjoern, Z. Floriano, C. Michele, On thesimulation of gridmarket coordination approaches, Journal of Grid Computing6 (2008) 349–366.
[52] P. Behnaz, B. Koen, V. Stamatis, A. Luc Onana, A dynamic pricing and biddingstrategy for autonomous agents in grids, in: 6th International Workshop onAgents and Peer-to-Peer Computing, AP2PC 2007, Honululu, Hawaii, USA,May14–18, 2007, Revised and Selected Papers, Springer-Verlag, 2007, pp. 55–71.
[53] T. Zhu, R.G. John, Market-based grid resource allocation using a stablecontinuous double auction, in: Proceedings of the 8th IEEE/ACM InternationalConference on Grid Computing, IEEE Computer Society, 2007.
[54] W. Chuliang, L. Minglu, L. Xinda, Grid resource management based oneconomic mechanisms, Journal of Supercomputing 42 (2007) 181–199.
[55] A. Marcos, R. Buyya, An evaluation of communication demand of auctionprotocols in grid environments, in: Proceedings of the 3rd InternationalWorkshop on Grid Economics and Business, GECON 2006, World ScientificPress, 2006.
[56] B. Pourebrahimi, K. Bertels, G.M. Kandru, S. Vassiliadis, Market-based resourceallocation in grids, in: Second IEEE International Conference on e-Science andGrid Computing, e-Science’06, 2006, p. 80.
[57] M. Placek, R. Buyya, Storage exchange: a global trading platform for storageservices, in: 12th International European Parallel Computing Conference(EuroPar), in: LNCS, Springer-Verlag, Berlin, Germany, 2006, pp. 425–436.
[58] E. Torsten, N. Dirk, R. Michael, S. Bjorn, S. Werner, V. Daniel, On the designof a two-tiered grid market structure, in: Proceedings of the MultikonferenzWirtschaftsinformatik, MKWI’06, Passau, Germany, 2006.
[59] K. Umesh, G. Daniel, Double auction protocols for resource allocation in grids,in: Proceedings of the International Conference on Information Technology:Coding and Computing, ITCC’05-Volume I, vol. 01, IEEE Computer Society,2005.
[60] G. Daniel, D. Anubhav, Auction-based resource allocation protocols in grids,in: In the Proceedings of the 16th IASTED International Conference on Paralleland Distributed Computing and Systems, PDCS 2004, ACTA Press, Cambridge,MA, 2004, pp. 20–27.
[61] J. Gomoluch, M. Schroeder, Market-based resource allocation for gridcomputing: a model and simulation, in: Proceedings of the First InternationalWorkshop on Middleware for Grid Computing, 2003.
[62] H. Minghua, L. Ho-fung, N.R. Jennings, A fuzzy-logic based bidding strategyfor autonomous agents in continuous double auctions, IEEE Transactions onKnowledge and Data Engineering 15 (2003) 1345–1363.
[63] X. Liu, L. Zhao, A method based on iterative combinatorial auction mechanismfor resource allocation in grid multi-agent systems, in: InternationalConference on Intelligent Human–Machine Systems and Cybernetics, 2009.IHMSC’09, 2009, pp. 36–39.
[64] Björn Schnizler, MACE: a multi-attribute combinatorial exchange, in: Negoti-ation, Auctions, and Market Engineering, Springer, Berlin, Heidelberg, 2008.
[65] S.E. Middleton, M. Surridge, S. Benkner, G. Engelbrecht, Quality of servicenegotiation for commercial medical grid services, Journal of Grid Computing5 (2007) 429–447.
[66] A. Attanasio, G. Ghiani, L. Grandinetti, F. Guerriero, Auction algorithms fordecentralized parallel machine scheduling, Parallel Computing 32 (2006)701–709.
[67] J. Brunelle, P. Hurst, J. Huth, L. Kang, C. Ng, D. Parkes, M. Seltzer, J. Shank,S. Youssef, EGG: an extensible and economics-inspired open grid computingplatform, in: 3rd International Workshop on Grid Economics and BusinessModels, GECON 2006, Singapore, 2006, pp. 140–150.
[68] A. Alvin, et al. Resource allocation in federated distributed computinginfrastructures, in: Proceedings of the 1stWorkshop on Operating System andArchitectural Support for the on-Demand IT InfraStructure, 2004.
[69] R. Subrata, A.Y. Zomaya, B. Landfeldt, Cooperative power-aware scheduling ingrid computing environments, Journal of Parallel and Distributed Computing70 (2009) 84–91.
[70] Z. Han, L. Xiaolin, Efficient grid task-bundle allocation using bargaining basedself-adaptive auction, in: Proceedings of the 2009 9th IEEE/ACM InternationalSymposium on Cluster Computing and the Grid, CCGrid 09, IEEE ComputerSociety, 2009, pp. 4–11.
[71] Q. Dang Minh, A. Jorn, Bilateral bargaining game and fuzzy logic in the systemhandling SLA-based workflow, in: Proceedings of the 22nd International Con-ference on Advanced Information Networking and Applications-Workshops,IEEE Computer Society, 2008.
[72] A. Marcos Dias de, R. Buyya, A cost-aware resource exchange mechanismfor load management across grids, in: Proceedings of the 2008 14th IEEEInternational Conference on Parallel and Distributed Systems, IEEE ComputerSociety, 2008.
[73] W. Junyan, X. Yuebin, L. Guanfeng, P. Zhenkuan, H. Yongsheng, A newresource discoverymechanismwith negotiate solution based on agent in grid
environments, in: Proceedings of the 2008 the 3rd International Conferenceon Grid and Pervasive Computing—Workshops, IEEE Computer Society, 2008.
[74] D.d.A. Marcos, S. Werner, E. Torsten, B. Rajkumar, Enabling the simulation ofservice-oriented computing and provisioning policies for autonomic utilitygrids, in: Proceedings of the 4th International Workshop on Grid Economicsand Business, GECON’07, Rennes, France, 2007.
[75] S. KwangMong, Relaxed-criteria G-negotiation for grid resource co-allocation,in: ACM SIGECOM: E-commerce Exchanges, vol. 6, 2007, pp. 37–46.
[76] L. Jiadao, R.R. Yahyapour, A strategic negotiation models for grid scheduling,International Transactions on Systems Science and Applications (2007)411–420.
[77] P. Ghosh, N. Roy, S.K. Das, K. Basu, A pricing strategy for job allocation inmobilegrids using a non-cooperative bargaining theory framework, Journal of Paralleland Distributed Computing 65 (2005) 1366–1383.
[78] K.M. SIM, A survey of bargaining models for grid resource allocation, ACMSIGecom Exchanges 5 (2005) 22–32.
[79] L. Kevin, et al., Tycoon: an implementation of a distributed, market-basedresource allocation system, Multiagent and Grid Systems (2005).
[80] X. León, T.A. Trinh, L. Navarro, Using economic regulation to prevent resourcecongestion in large-scale shared infrastructures, Future Generation ComputerSystems 26 (2010) 599–607.
[81] J.N. Falavinha, A. Manacero, M. Livny, D. Bradley, The owner share schedulerfor a distributed system, in: Parallel Processing Workshops, 2009, ICPPW’09,International Conference on, 2009, pp. 298–305.
[82] C. Li, L. Li, Competitive proportional resource allocation policy for computa-tional grid, Future Generation Computer Systems 20 (2004) 1041–1054.
[83] J. Sherwani, et al., Libra: a computational economy-based job schedulingsystem for clusters, Journal of Software: Practice and Experience 34 (2004)573–590.
[84] B.N. Chun, P. Buonadonna, A. AuYoung, N. Chaki, D.C. Parkes, J. Shneidman, A.C.Snoeren, A. Vahdat, Mirage: a microeconomic resource allocation system forsensornet testbeds, in: The Second IEEE Workshop on Embedded NetworkedSensors, 2005. EmNetS-II, 2005, pp. 19–28.
[85] C. Isaac, S. Ramon, A. Oscar, J. Liviu, F.R. Omer, Optimizing decentralisedgrid markets through group selection, International Journal of Web and GridServices 4 (2009) 357–366.
[86] C. Massimiliano, G. Stefano, Resource allocation in grid computing: aneconomic model, WSEAS Transactions on Computer Research 3 (2008) 19–27.
[87] G. Kunal, G. Arobinda, Resource selection in grids using contract net,in: Proceedings of the 16th Euromicro Conference on Parallel, Distributed andNetwork-Based Processing, PDP 2008, IEEE Computer Society, 2008.
[88] A. Di Stefano, C. Santoro, An economic model for resource managementin a grid-based content distribution network, Future Generation ComputerSystems 24 (2008) 202–212.
[89] R. Rajiv, H. Aaron, B. Rajkumar, SLA-based cooperative superschedulingalgorithms for computational grids, ACM Transactions on Autonomous andAdaptive Systems (2008).
[90] M. Dominiak, M. Ganzha, M. Gawinecki, W. Kuranowski, M. Paprzycki,S. Margenov, I. Lirkov, Utilizing agent teams in grid resource brokering,Multiagent and Grid Systems (2007).
[91] S. Paurobally, A framework for web service negotiation, ACM Transactions onAutonomous and Adaptive Systems 2 (2007) 14.
[92] D. Ouelhadj, J. Garibaldi, J. MacLaren, R. Sakellariou, K. Krishnakumar, A multi-agent infrastructure and a service level agreement negotiation protocol forrobust scheduling in grid computing, in: Advances in Grid Computing—EGC2005, Springer-Verlag, 2005, pp. 651–660.
[93] Z. Juhasz, P. Paul, Scalability analysis of the contract net protocol, in: ClusterComputing and the Grid, 2002, 2nd IEEE/ACM International Symposium on,2002, pp. 346–346.
[94] S. Pooja, M. Minakshi, Sunaina, Multi-agent systems for adaptive and efficientjob scheduling service in grids, International Journal of Computer Applications1 (2010) 19–23.
[95] C. Giorgos, K. Chris, R. Buyya, Grid economics: 10 lessons from finance, JointTechnical Report GRIDS-TR-2003-3, IBM Research Zurich and Grid Computingand Distributed Systems Laboratory and University of Melbourne, 2003.
[96] I. Foster, C. Kesselman, The Grid Blueprint for a Future ComputingInfrastructure, Morgan Kaufmann, 1999.
[97] I. Foster, Z. Yong, I. Raicu, S. Lu, Cloud computing and grid computing 360-degree compared, in: Grid Computing EnvironmentsWorkshop, 2008. GCE’08,2008, pp. 1–10.
[98] A.Michael, F. Armando, G. Rean, D.J. Anthony, H.K. Randy, K. Andrew, L. Gunho,A.P. David, R. Ariel, Z. Matei, Above the clouds: a berkeley view of cloudcomputing, 2009.
AminulHaque is currently studyinghis Ph.D. degree at theSchool of IT, Monash University, Malaysia. He completedhis B.Sc. in Physics from Shah-Jalal University of Scienceand Technology, Bangladesh. His area of research interestsare: Grid computing and Agent technology.
A. Haque et al. / Future Generation Computer Systems 27 (2011) 1056–1069 1069
Saadat M. Alhashmi is a Senior Lecturer at the Schoolof Information Technology, Monash University, Malaysia.He holds a B.Sc. in Engineering from AMU, India, anM.Sc. in Automatic Controls and Systems Engineeringfrom The University of Sheffield and a Ph.D. fromSheffield Hallam University, UK. His area of researchinterests are: Supply ChainManagement, Fuzzy Logic, GridComputing, Multimedia Information Retrieval and Multi-agent Methodology.
Rajendran Parthiban is a Senior Lecturer at the School ofEngineering, Monash University, Malaysia. He completedhis BE (Hons) in 1997 and Ph.D. in 2004 both fromUniversity of Melbourne, Australia. His research andprofessional interests are in the cost comparisons ofvarious optical network architectures, energy calculationsof future optical networks, security and localization inwireless sensor networks, cost-effective applications ofRadio Frequency Identification (RFID) andGrid computing.