Taxonomy and issues for antifragile-based multimedia cloud computing

13 Pages·2016·1.45 MB·English

by  Syed Fawad Haider

JReliableIntellEnviron(2016)2:37–49
DOI10.1007/s40860-016-0017-7
REVIEW
Taxonomy and issues for antifragile-based multimedia cloud
computing
SyedFawadHaider1 · LaraibAbbas1 · AmjadAli2 · MuddesarIqbal3 ·
ImranRaza2 · SyedAsadHussain2 · DougYoungSuh4
Received:19August2015/Accepted:6February2016/Publishedonline:27February2016
©SpringerInternationalPublishingSwitzerland2016
Abstract Cloud computing has become one of the most ment of MCC has given rise to various challenges related
dynamic and adoptable computing paradigms. Multimedia to resource allocation and task management. Antifragility
CloudComputing(MCC)isoneoftoday’shotresearchtopic. is a key to such environments, to let the disorder drive the
MCC is proven to be a most dynamic and efficient plat- strengthofthesesystems.Thispaperismainlydividedinto
form for managing a large amount of multimedia contents three parts. The first part discusses in detail the available
with maximum deployment of computing and processing state-of-the-artrelatedtoresourcemanagementunderMCC.
resourcesattheserviceproviderinsteadofusers.Resilience Similarly, the second part presents the comprehensive lit-
and dependability are two key constituents to assure the erature review on the task management in MCC. The third
reliability and availability of any service in the presence part presents the critical analysis and open research issues
of errors and system failures. The heterogeneous environ- in MCC which help the researcher to define their research
objectivesinthefieldofMCC.
B
MuddesarIqbal
[email protected] Keywords Cloud computing · Quality of service ·
Antifragility·Qualityofexperience·Resourceallocation
SyedFawadHaider
[email protected]
LaraibAbbas
[email protected]
1 Introduction
AmjadAli
[email protected]
In the current era of technology, multimedia devices are
ImranRaza
[email protected] evolving exponentially. Such a rapid growth of multime-
dia devices has introduced a larger number of multimedia
SyedAsadHussain
[email protected] applications and services. Thus, multimedia users over the
globe are producing a huge amount of multimedia con-
DougYoungSuh
[email protected] tentswhichrequirehighprocessingandstoragecapabilities.
Hence, processing of such a huge amount of multimedia
1 FacultyofComputingandInformationTechnology,
contents in timely and efficient manner by fulfilling their
UniversityofGujrat,Gujrat,Pakistan
users’ Quality of Service (QoS) requirements is a highly
2 DepartmentofComputerScience,Communicationand
challenging [1–3] task. Moreover, due to diverse nature of
NetworksResearchCentre,COMSATSInstituteof
multimedia data, anomaly management is also one of the
InformationTechnology,Lahore,Pakistan
bigchallenges.Thus,MultimediaCloudComputing(MCC)
3 Pak-UKInstituteofInnovativeTechnologiesforDisaster
paradigm, which is motivated by the heterogeneous nature
Management,UniversityofGujrat,Gujrat,Pakistan
ofmultimediaapplicationandservices,isdifferentthanthe
4 DepartmentofElectronicsandRadioEngineering,Collegeof
othergeneralpurposecloudcomputingparadigms[4,5]and
ElectronicsandInformation,KyungHeeUniversity,Yongin
446-701,SouthKorea hascapabilitiestoresolvesuchissuesandchallenges[6–8].
123
38 JReliableIntellEnviron(2016)2:37–49
Cloud computing is a large-scale distributed paradigm a simple scenario. The authors interpreted resilience as the
where users can access various application softwares and emergingresultofadynamicprocessinwhichprocessrepre-
infrastructure through the Internet as shown in Fig. 1. The sentsthedynamicinterplaybetweenthebehavioursexercised
service-orientedarchitectureandelasticityofinfrastructure by a system and those of the environment it is set to oper-
providedbycloudcomputingisoneofthemajorattractions ate in. Similarly, in [18] the authors proposed a model of
fororganizationstooptitfortheirservices.Cloudcomput- thefidelityofopensystems.Theauthorsinterpretedfidelity
ing uses a enhanced optimization and focuses on sharing as the compliance between corresponding figures of inter-
resources among the cloud users to reduce the overall cap- estintwoseparatebutcommunicatingdomains.In[19]the
italinvestmentaswellasoperatingcosts[9–12].Userscan authorsdiscussedandpresentedseveralexamplesonongo-
benefit from flexible service models of cloud computing ingresearchtoemploytheconceptthatinsteadofdesigning
which offer Infrastructure as a Service (IaaS), Platform as systems to meet known requirements which always lead to
aService(PaaS),andSoftwareasaService(SaaS),accord- fragilesystemsatsomedegree,systemsshouldbedesigned
ingtotheirdemandsandrequirements[9,13,14].However, whereverpossibletobeantifragile.
failure is unavoidable in cloud architecture due to internal Cloudcomputingisthemostadoptablefieldofcomputing
or external conditions such as human-made faults, unre- worldwhichiscategorizedintothreedifferenttypes:Public,
liable hardware/software, and natural disasters. Therefore, Private,andHybridclouds.Apubliccloudisdeployedbya
the idea of using these failures to improve the resilience thirdpartyandprovidesservicestotheendusersondemands
and maintainability of cloud environment is the basic con- such as Google and Amazon. Private cloud is deployed by
cept of antifragility [15]. Taleb introduced the concept of acompanyforitinternaluse.Similarly,InHybridcloud,a
antifragilitystatingthat“Antifragile:Thingsthatgainfrom company stores some of its data on public cloud and most
disorder”[16].Thisapproachcanbeusedinacloudenviron- secret data are stored on its own private cloud [20]. In the
menttoincreaseitsflexibilityandproductivityusingdifferent next subsection, we present a comparison study over both
techniques. The authors in [17] introduced and discussed a computingparadigms.
frameworkforresilienceofcomputational systemsthrough
Fig. 1 Generalcloud
computingarchitecture
123
JReliableIntellEnviron(2016)2:37–49 39
1.1 Multimediacloudvs.conventionalcloudcomputing ing. Thus, MCC must contain high-performance Graphical
paradigm ProcessingUnits(GPUs),CentralProcessingUnits(CPUs),
andstoragecapacitiesashowninFig.2toperformtheeffi-
Traditional Cloud Computing (TCC) architecture refers cient and timely processing of multimedia contents [24].
ubiquitous, convenient, on-demand network access to a However, TCC and MCC support a few common services
shared pool of configurable computing resources such as andapplicationsbutgenerallytheirarchitecturesarewidely
servers, networks, storage, applications and services. TCC different from each other. Moreover, multimedia applica-
works as a utility computing where users employ services tions and services require different types and capacities of
andpayforwhattheyuse.AmazonEC2,MicrosoftAzure, processing and storage which motivates to design a cloud
andGoogleAppEnginearefewexamplesofPublicclouds. that can handle all issues related to multimedia application
TCChandlessomekeychallengeslikescalability,QoS,and andservice-relatedissues.Thefollowingaretwomaintypes
virtualizationwhichmakestheTCCanefficientcomputing ofMCC.
paradigm[21,22]. Cloud-aware multimedia (Cloud Media) Cloud media [24]
Duetoavailabilityofhighcomputingandstorageresour- isdefinedas“Multimediaservicesandapplicationssuchas
cesatlowcost,cloudisanefficientandoptimaloptionfor storage, sharing, authoring, mash-up, rendering, retrieval,
processingandstoringahugeamountsofdata[23].However, adaptation, and delivery can efficiently and effectively uti-
duetoheterogeneousnatureofmultimediadevices,applica- lize cloud resources to enhance the Quality of Experience
tion,services,anddifferentQoSrequirementsofmultimedia (QoE) of multimedia users”. The main characteristics are
traffics,theTCCisnotanappropriatecomputingparadigm. listedbelow:
Hence,anewefficientandfastcomputingparadigm,called
MultimediaCloudComputing(MCC),hasbeenintroduced, 1. Storage is always made available to the cloud users so
which processes multimedia contents (i.e., images, videos, theycaneasilysharetheirdataeverywhere.
and graphics, etc.) in a distributed manner and eliminates 2. Cloudcanbeusedefficientlytoeditdifferentsegments
the installations of media applications on local machines ofmultimediacontentsandcombinethem.
[24]. Therefore, MCC is quite different than TCC due to 3. Cloud system can be used efficiently and speedily for
timely processing and strict QoS requirements of multime- renderingandretrievalofmultimediacontents.
diacontents.Multimediacontentsrequirealargeamountof 4. Cloudcanbeusedtotransformdifferentmediacontents
computing resources for their efficient and timely process- anddeliverthemtotheusers.
Fig. 2 MCCarchitecture
123
40 JReliableIntellEnviron(2016)2:37–49
Fig. 4 Queuingmodel
based on queuing model. There are certain issues such as
costoptimization,serviceresponsetime,QoE,resourceallo-
cation cost that severely affect the performance of MCC.
Thus,theserviceresponsetime,thatisatimebetweenarrival
of user request and its departure, should be minimized to
Fig. 3 TaxonomyofMCC
enhancetheQoEofthemultimediauser.
Single-class/multi-class service Nan et al. [26] also intro-
ducedanoptimalresourceallocationschemeforMCCbased
Multimedia-aware cloud (Media Cloud) Media cloud is a
on priority services scheme in both single-class and multi-
newmultimediacomputingparadigm[24]thatensuresmul-
class services case. In their proposed scheme, cloud is
timedia features such as QoS and supports various types
designed in the form of a data centre which consists of a
of multimedia contents and heterogeneous devices. Many
master server and group of computing servers. The master
solutionsarediscussedfortask,resource,andQoSmanage-
serveractslikeacontrollernodethatreceives,schedules,and
mentinMediaCloud[24].ToenhancetheQoS,storageand
forwardsuserrequeststotheothercomputingserverswhich
processingofmultimediacontentsareperformedbycloud.
furtherprocesstheseuserrequests.Themasterserverandthe
In the following sections, we present a comprehensive
computingserversformalogicaltree,inwhichmasterserver
review of the state-of-the-art related to MCC. This paper
actslikearootnodeandcomputingserversaretheleafnodes
mainly focuses on two important issues: (1) task and (2)
connectedwiththeroot.Schedulerateandcomputationrates
resourcemanagement inMCC.Thecomplete taxonomy of
areusedtoassigntheweightstothelinksbetweentheroot
MCCispresentedinFig.3.
andleafnodes,respectively.
Priority services scheme mainly contains three types of
queues; (1) schedule queues, (2) computation queues, and
2 Resourcemanagementinmultimediacloud (3) transmission queues as shown in Fig. 4. The schedule
computing queue, installed at master server is responsible for receiv-
ing and scheduling user’s requests for further processing.
Inthissection,wediscussacomprehensiveoverviewofdif- Afterscheduling,theusertasksreachthecomputationqueues
ferenttechniquesformultimediacloudcomputingunderthe of their respective server and wait for execution. When
category of resource management in MCC. MCC is newly therequests/tasksareprocessedincomputationserverthey
emerged computing paradigm and it’s highly challenging are forwarded to the transmission server, where they wait
becausetheMCCusersrequiredhighlyefficientandtimely in transmission queue before they are transmitted to their
computation of multimedia data which makes it different respective users as shown in Fig. 5. Media cloud can opti-
fromconventionalcloudcomputing.ResourceManagement mally process multimedia applications and services while
in terms of processing and storage is a key issue in MCC ensuring their respective users QoS requirements. A num-
asdifferentmultimediacontentsrequiredifferentcomputing berofmultimediaapplications,i.e.images/videosprocessing
resources. and3Drenderingcannotefficientlyprocessclientmachines
Queuing model Quality of Experience (QoE) is one of the as they need intensive storage and processing capacity, so
main objectives of a media cloud provider. Nan et al. [25] theseapplicationsefficientlyprocessinMCC.Therearetwo
introducedanoptimalresourceallocationschemeforMCC majorchallenges;(1)serviceresponsetimeand(2)resource
123
JReliableIntellEnviron(2016)2:37–49 41
the capacity enhancement of media servers and parameters
required to enhance the QoS in multimedia cloud. Energy
consumption,resourceallocation,cost,andcompletiontime
shouldbeconsideredbeforeallocatingresourcestothemedia
tasksinMCC.Theirproposedresourceallocationschemeis
dividedintothreesections;(1)taskanalysis,(2)cloudbroker,
and(3)resourcemanager.Whenauserrequestarrivesatthe
cloud, it is analysed to figure out its QoS parameters. The
cloudbrokerthencomparestheQoSrequirementsofuser’s
requestwiththecloudresourcesandestimatestherequired
resourcesforthegivenjob.Finallytheresourceserversassign
thesetaskstotheactualvirtualmachineswhichprocessuser’s
requests.
Authors introduced a utility function in their proposed
Fig. 5 Priorityservices-basedqueuingmodel schemethatisbasedonthegametheoryinwhichtheyeffi-
cientlydistributethecloudresourcesamongallcloudusers.
Theutilityfunctioncoverstheobjectivesofboththeusersand
costoptimizationthatcanseverelyaffecttheperformanceof cloudproviders.Theusersareconcernedwiththemultime-
MCC. diataskcompletiontimeandcostofrequiredresourcesand
DynamicresourceallocationMediacloudenhancestheper- thecloudprovider’sconcernsareaboutenergyconsumption
formanceofmediaprocessingasitworkedindynamic,dis- inthecloudplatform.Themainobjectiveofutilityfunctionis
tributed,parallel,andsharedenvironment[27].Theresource tofulfiluserandcloudproviderconcernsandultimatelygain
allocationformediatasksinsuchadynamicandsharedenvi- user’ssatisfactionandminimizetheenergyconsumptionin
ronmentisabigchallengetoachievesystemefficiencyand thecloudplatform.Thisutilityfunctionallocatesresourcesto
QoSforthemediausers.Manypreviouslypresenteddesigns themediataskintwophases.Inthefirstphase,arrivedmedia
for media cloud have strived to improve system efficiency tasks occupy the cloud resources without considering the
but could not enhance theQoS for media tasks.Authors in sharednatureoftheseresources.Iteventuallycreatesacon-
[27]investigatedthepropertiesofmediatasksandrequired flictamongothermediatasks,whichiseventuallyresolved
resourcestoprocesstheminamediacloud. by reallocating the cloud resources to optimize the perfor-
The main research focus of work presented in [27] is to manceofMCC.
improvethesystemefficiencyandQoSbyfindingandassign- MediataskQoSbasedresourceallocationBohaietal.[29]
ing exact amount of resources for any media task. For this discussed the overall QoS for media task and resource uti-
purpose,adynamicallocationschemeisintroducedbasedon lizationinMCCandproposedanefficientresourceallocation
machine learning and previous task history. Survival func- scheme based on media task QoS for MCC. The computa-
tion controls the resource allocation and grants the QoS to tional weight of the task which is found to fulfils the QoS
media tasks. This dynamic scheme is compared with the weight of media task is measured in terms of QoS weight
static resource allocation schemes and is found to be the vector and an expected resource vector. Later, the resource
most appropriate scheme for multimedia cloud computing. similarityvectorfromexpectedresourcesiscalculatedalong
Moreover,theirproposedschemealsoaddressestheresource withanalternativeresourcevectorbyusinglinearnormaliza-
allocation problem in cloud based media platform. Media tion.Attheend,servicesatisfactionbyEuclideandistanceis
tasksthatarriveatmultimediacloudareheterogeneousand formulizedandresourcesareallocatedaccordingtotheser-
have different QoS requirements. They initially proposed a vicesatisfaction.Theirproposedarchitectureisdividedinto
static resource allocation scheme, in which resources are threecomponents;(1)taskpool,(2)mediaservicemanager,
stored in a pool and are assigned to tasks as they arrive. and(3)cloudplatform.
Media task requirements in terms of processing and stor- The task pool is the component where media tasks, i.e.
age vary quite often, so a machine learning based dynamic video, images, and graphics are processed. Media Service
resourceallocationschemehasbeenproposedthatpredicts Manager (MSM) is responsible for task analysis, schedul-
theexactamountofresourcesandenhancestheperformance ing,and resourceallocation andassignsappropriate virtual
ofmediaprocessing. machine to the media task. The MSM is a coordinator
QoS-basedresourceallocationYiruietal.[28]discussedthe between the task pool and cloud platform. The cloud plat-
resource allocation problem in MCC and proposed a QoS- form is the actual cloud resource that process media tasks.
based efficient resource allocation scheme to enhance the As the allocation of resources to the media task is based
performanceofMCC.Theirproposedschemeconsidersboth onQoSweightvectorssothetaskanalysismodulemustbe
123
42 JReliableIntellEnviron(2016)2:37–49
aware of all the physical resources. It can be easily com- resourcesandstreamingservicestotheusers.Theirproposed
puted based on the expected resource vector. Furthermore, solutionconsidersthetime-discountandnon-lineartariffthat
thereisheterogeneityinmediatasksandcloudplatformso is changed by a cloud provider for the purpose of resource
allocationandschedulingofresourcesisbasedonexpected reservation in cloud. On the basis of prediction of stream-
resources vector. MSM selects appropriate resources from ing capacity’s future demand, the financial cost of Media
resourcepoolandsendsresourceinformationtothevirtual CloudProvider(MCP)isminimizedbyusinganalgorithm
machine (VM) module. A VM is created which processes that accumulates the reservation time and amount of cloud
themediataskandreturntheresourceswhenfinished. resources.Theirproposedschemealsoensuresthatenough
Cost effective resource allocation MCC computing para- resourcesarebeingreservedincloudwithoutanywastage.
digmisalsoveryeffectiveandefficientsolutionforE-health Resourceallocationforcloud-basedvideosurveillanceplat-
systems [30]. E-health systems require high capacity of form Authorsin[32]presentedaresourceallocationscheme
computing resources, i.e. high processing power, storage forcloud-basedvideosurveillanceenvironment.Theirpro-
capacity, and network bandwidth. Keeping in view the posed scheme focuses on optimizing the VM resources to
dynamicanddelay-sensitivenatureofE-healthapplications, fulfil various types of user services which are provided by
the cloud computing is an optimal option because it can MCC. Single service is not effective for user requests but
dynamically adjusts the resources for such systems. How- compositeservicescouldbeefficientandoptimal.Theirpro-
ever,issueslikecosteffectivenessofresourceallocationand posed design is the composition of two mechanisms: (1)
energy consumption can still affect the performance for E- LinearProgrammingModeland(2)HeuristicApproachthat
health systems. The authors in [30] focus on these issues candynamicallyallocateresourcesinmediacloud.Costopti-
andproposeacosteffectiveQoS-awareresourceallocation mization and service response time are being improved by
scheme for MCC based E-health systems. Their proposed thisapproach.Theytestedtheirproposedprototypeinitially
resourceallocationschemeovercomestwoimportantgoals. insidetheAmazoncloudunderalimitedtypeofextent.The
Thefirstistominimizetheoverheadincloudplatformand VSS directory is a composition for the users such as fire
guarantees the required resources for VM and given media fightersandsecuritypersonnelthatprovidethepaymentser-
tasks to finish timely. The second is to minimize the cost vices,analysis,sharing,streaming,andtranscodingservices
andenergyconsumptiontohandleallVMswithlessnumber throughaninterfaceofwebbrowser.
of servers. The servers that are not in use should be pow- Media-edgecloud Wenwuetetal.in[23]introducedamedia-
ered off. To minimize task completion time and overhead edgecloud(MEC)designforMCC.Theirproposeddesign
onservers,thereshouldbeatradeoffbetweenthecostand providesaparallelanddistributedprocessingandQualityof
energyconsumptionviaincreasingordecreasingtheutiliza- Serviceschemes.Themediacloudisdividedintothreeclus-
tionofservers. ters;(1)CentralProcessingUnit(CPU),(2)Storage,and(3)
Their proposed idea is based on the Nash Bargaining GraphicsProcessingUnit(GPU).Mediagraphicprocesses
Gamethatcontainscommoditiesandbargainingasagame areperformedonGPUclusters.However,theCPUclusters
theoreticapproach.Intheirproposedcloudresourcealloca- performs the general media processes and storage is man-
tionscheme,theentiredatacentreactsasbargainingmarket agedbystorageclusters.Amediacloudproxyisalsoused
andtheVMswhichprocessthemediatasksactascommodi- tohandletheheterogeneousmediacontentsofmobileusers.
ties in the bargaining market. All the servers participate in Resource allocation controller for cloud-based adaptive
the market and bargain for their desired commodities. The videostreaming Lucaetal.[33]presentedaresourcealloca-
Nashbargaininggameisresponsibleforassigningthedesired tioncontrollerforvideostreaminginMCC.Theaimwasto
commoditiestotheserversinthemarketinsuchawaythat providehigh-qualityvideostreamingtoendusersatminimal
the social welfare is achieved and any VMs do not exceed distributioncost.Acontrollerisresponsibleforresourceallo-
theircapacities. cationthatdynamicallyarchivestheresourcemanagement.
Resource allocation for media streaming applications Amr Thisresourceallocationcontrollercontainedthreemodules:
et al. [31] proposed a prediction based resource allocation (1)LoadBalancer,(2)Resourceallocation,and(3)Stream
scheme for media streaming applications. Their proposed switching adaptation controller (SSAC). First two modules
schememainlyfocusesonoptimizingthetariffcostofmedia are synchronized such that the user sends a request which
cloudbyminimizingthemediacloudreservedresources.The isassignedtotheactive serverbyaloadbalancer andthen
wholemediacloudisdividedintothreemodules;(1)demand adaptive video streaming session is started. SSAC instance
forecasting that predicts the future streaming needs from is also started with this process. In the mean while, active
usersbasedonprevioususagepatterns,(2)cloudbrokerthat serversendsfeedbacktoRACthroughmonitorthatdecides
allocatestherequiredresourcesandreservesafewothersfor the behaviour of the machine to be ON or OFF. Figure 6
specifictimeandimplementsthepredictionschemeobtained describes this design. To evaluate the performance of the
inforecastingmoduleand(3)mediaproviderthatprovides proposed solution, authors used CDNSim after some mod-
123
JReliableIntellEnviron(2016)2:37–49 43
System (NFS) is used to mount the Android File System
(AFS).TomountthecatalogoftheAFSonthelocalserver,
TCP/IP protocol stack is being used by NFS. Application
filescanberemoved,added,ormodifiedbytheuserthrough
asingleremotecontrol.Videofilesareusedtotesttheperfor-
mance of the hardware platform. The performance of DSP
chip, ARM chip, and mixed ARM is compared on system
operationstate.
QoS-aware data forwarding for multimedia streaming ser-
vice Seokhoonetal.[23]presentedaQoS-awareschemefor
dataforwardingforMultimediaStreamingServices(MSS).
Intheproposedscheme,theQDFAsynchronizationprocess
isusedtoimprovethetransmissionefficiencywiththeappli-
cationofmoreaccurateandprecisemeasuresoftransmission
time under IEEE 1588 standard. To provide better quality,
an improved and novel scheme, based on QDFA, uses the
techniques of GMPLS, DiffServ, POSIA, and IntServ. The
characteristicsofbackgroundtrafficlikeSMTPandinterac-
Fig. 6 Systemoverview tivetrafficlikeHTTPisusedtobursttheQDFA.Toevaluate
the performance of proposed algorithm, different schemes
ifications on the bases of INET and Omnet++ framework. suchIntServ,MPLS,andDiffServarecomparedwithQDFA.
Theproposedschemeused250destinationsforeachsideand
Thefollowingmetricsarebeingmeasuredforthepurposeof
multimediastreamingtrafficisusedinsimulationasatraffic
performanceevaluationofresourceallocatorcontroller:
typeforP2Phybridnetwork.
Two-stage approach for task and resource management
(cid:129) Thenumberofactivemachines.
Biao et al. [35] presented a two-stage scheme for task
(cid:129) ThecostofCPUusage.
andresourcemanagementinMCC.Theproposedapproach
(cid:129) Thefractionofstreamsthatobtainthemaximumvideo
mainlyfocusesonresourcemanagement,inwhichitdefines
level.
the way of assigning VMs to the actual servers and task
management where VMs were assigned the tasks. Their
Thesimulationisdonewiththehelpofrealisticworkload. proposed heuristic scheme for resource management and
The results are taken considering the delivery cost so that taskmanagementhasbeendonebyqueuingmechanismvia
their proposed architecture can provide high-quality video addingadeadlineapproachinuserrequests.Then,taskman-
streamingtotheenduserswhilesavingthedistributioncosts. agerandresourceallocatoroptimallyminimizethecostand
Powersavingschemeformultimediastreamingservices Yi- enhance the QoS for multimedia services. The following
Wei et al. [34] introduced a power saving based scheme Fig.7describesthedesignofthisscheme.
for multimedia streaming services in MCC. The authors
proposed a multimedia service architecture for streaming
services over embedded system that uses the digital signal
processor.Themainfocusisondeliveringhigh-qualitymedia 3 Taskmanagementinmultimediacloud
serviceincloudandsavethepowerformobiledevices.The computing
wholedesignisdividedintothreeparts;(1)AndroidOSon
hardwareplatform,(2)Powersavingscheme,and(3)Digi- Inthissection,wediscussacomprehensiveoverviewofdif-
talsignalprocessing(DSP)module.Whentheuserrequests ferenttechniquesformultimediacloudcomputingunderthe
forvideostreaming,theserverloadsthedesiredvideointo categoryoftaskmanagementinMCC.
memory buffer using HTTP. The multimedia framework is Effectiveloadbalancingforcloudbasedmultimediasystem
calledbyapplicationlayertoanalysevideoformatandeven- Huietal.[36]highlightedakeychallengethatcanseverely
tuallythepacketsaresenttotheDSPfordecoding.Finally, affecttheperformanceofthecloudcomputing(i.e.,assign-
thedesiredvideopacketsaresenttothememorybufferand ingaccurateamountofcloudresourcesinashortperiodof
users,respectively,throughadvancedARMLinuxhardware timetothemultimediaapplications).Toovercomethisissue
platform. anefficientloadbalancingschemeforcloud-basedprocess-
Forachievinganefficientembeddedsystemdevelopment ingisintroduced,whichiscalledacloud-basedmultimedia
and to minimize the kernel burning time, the Network File load balancing scheme. This scheme manages the capacity
123
44 JReliableIntellEnviron(2016)2:37–49
Fig. 7 Systemoverview
Fig. 9 Workloadschedulingformultimediacloudcomputing
produce large-scale workload. So, there is need of an effi-
Fig. 8 WorkloadschedulingforMCCarchitecture cient workload scheme that can enhance the performance
ofcloudcomputing.Differentworkloadschedulingschemes
ofindividualnodeinthecloudaswellasassignstheexact arediscussedin[37].Therearetwomainchallengeswhich
capacitytothecomingtrafficovertheparallellinks. can affect the performance of multimedia applications in a
The system consists of multiple scattered service nodes cloud platform; (1) service response time and (2) resource
whichprocessmultimediaapplicationsandservicesoverthe costminimization.Theauthorsin[37]addressedthesetwo
Internet. The proposed approach is divided into three dif- importantissuesandproposedagreedyalgorithmforwork-
ferent layers: (1) resource manager that is responsible for loadschedulinginMCC.
assigningappropriateresources;(2)nodemanager,whichis The proposed approach dynamically adjusts the work-
responsible for managing node in the system and allocates load in MCC. As the workload is varying with the time,
thetasks,and(3)nodeserversthatareresponsibleforexe- the time domain is divided into two different slots and
cuting multimedia tasks as presented in Fig. 8. In order to distributed between these two time slots. The VMs are
achieverequiredobjectivethemultimediataskisprocessed dividedintodifferentclassesandeachclassisdedicatedfor
inthisthree-layersystem.Intheproposedcloud-basedmul- a specific task, also known as virtual cluster, as presented
timedialoadbalancingscheme,theresourcemanagerselects is Fig. 9. Their proposed workload scheme addressed the
required and appropriate node for media tasks that depend serviceresponsetimeandresourcecostoptimizationbyfor-
ontheneedofusermediatasks.Finally,thenodemanager mulatingtheiroptimizedsolutions.Serviceresponsetimeis
assignsthatservicesnodetothemediatasks. usedasimportantfactortominimizethemeanresponsetime
Workloadschedulingformultimediacloudcomputing Mul- byoptimizingtheworkloadschedulingweightsfordifferent
timedia applications and services are time oriented and virtualclusters,subjecttothequeuingstabilityconstraintat
123
JReliableIntellEnviron(2016)2:37–49 45
each cluster. The workload conserving constraint, schedul- multimedia application, a distributed multimedia stream-
ing weight constraints, and resource cost optimization are ing system is proposed in [39]. The prime objective of
allaboutminimizingthetotalresourcecostbyjointlyopti- proposed system is to convert the variety of videos in
mizing the workload assignments and the allocated VMs MPEG-4 format which is used to further distribute them
in the on-demand reservation schemes. The optimization among a variety of heterogeneous multimedia users. Their
is carried out with subject to the application response time proposed scheme reduces the transcoding time by using
constraint,queuingstabilityconstraint,VMreservationcon- Hadoop file system and minimizes the content delivery
straint, workload conserving constraint, and the workload delays by using streaming distribution algorithm. Hadoop
schedulingweightconstraint. clustering techniques are used to enhance the performance
DynamicmultiserviceloadbalancingChun-Chengetal.[37] andefficiencyofcontentdeliveryusingHadooppoliciesand
presentedahierarchicalcloud-basedmultimediasystemthat strategies.
consisted of a resource manager, cluster heads, and server Their proposed system architecture is divided into three
clusters. User requests arrive at resource manager which modules; (1) the Hadoop-based distributed multimedia
transfersthemtothecontendedservercluster.Clusterhead transcodingmodule(HadoopDMT),(2)Hadoop-baseddis-
assigns the appropriate resources to the user requests. The tributed multimedia streaming module (HadoopDMS), and
mainissueaddressedinthiscloud-basedmultimediasystem (3) cloud multimedia management module (CMM). The
ishowtooptimallydistributeworkloadinthesystemwith- HadoopDMTtranscodesthevideointostandardformatsuch
outaffectingtheperformanceofthemultimediasystem.To as MPEG-4 which is compatible with most of the media
enhancetheworkloaddistributioninthissystem,aGenetic devices. When a video was transcoded, it moves to the
algorithm-baseddynamicmulti-serviceloadbalancingalgo- Hadoop distributed file system (HDFS) which is a part of
rithmisproposed. HadoopDMS. Then, HadoopDMS divides the media con-
Authorsin[36]proposedaschemethatisbasedonscheme tentsintosubpartsanddistributestheminthesystem.These
presentedin[38].Theirproposedschemehandlesthesimilar subpartsofmediacontentsaresavedonthreenodes,so,the
typesofmultimediatasksbyformulatingthedynamiccloud datacouldbemadeavailableincaseofanyfailure.Theprime
multimediasystem(CMS)accordingtotime.Timeisfurther purposeofCMMsystemistomanagetasks,scheduletasks,
dividedintomultipletimestepswhereineachtimestep,CMS andloadbalancing.
ismodeledasacompleteweightedgraph.CMScomponents Adaptive multimedia cloud computing centre applied on
arerepresentedintheformofweightedgraphU,inwhicha H.264/SVC streaming Multimedia applications have now
setofverticesrepresentsaserverclusters.Visasetofvertices become highly efficient and mature with the advancement
whichrepresentstheusersrequestsandsimilarly,Epresents innetworkbandwidthandinternettechnologies[40].Media
the set of edges between U and V. The proposed functions streaming has become a challenging issue in cloud-based
limitthemediaserversforaspecificmediataskbyproviding media system. The authors in [40] addressed these issues
theQoStothesystemandminimizingthelinkcostsbetween and proposed an adoptive multimedia system based on
clientandcloudserverclusters.Furthermore,thefunctional- H.264/SVC.Theirproposedschemeaddressedthecommuni-
ityofGeneticAlgorithm(GA)worksisbasedonevolutional cationlinksbetweentheclientsideandcloudprovideraswell
theoryofpopulation.GAfindstheoptimalsolutionfromthe astheloadbalancinginthecloudplatform.Algorithmin[39]
givensolutionset.GAalsofindsanefficientloadbalancing determinesthestreamingpathbetweentheclientandcloud
technique for CMS by dividing the problem into different provider based on the client side bandwidth and process-
steps. First, it randomly generates a set of all the possible ingpowertoprovidethehigh-qualityvideostreamingtothe
solutions. Second, it selects the optimal pair of solutions. clients.
Third, it compares the selected pair of optimal solution set The media cloud proposed architecture is comprised of
withtheoptimalsolutions.Basedonthisideatheproposed three different types of nodes: (1) index node, (2) content
dynamicloadbalancingschemeisdividedintofourcompo- node,and(3)streamingnodes. Theworkingofthismedia
nents;(1)initialization,(2)selectioncrossover,(3)mutation, cloudarchitectureisdescribedasfollows:
and(4)repair.Thesecomponentsworkinthesamewayas
GAperformsitsactions.
Cloud-based service architecture for multimedia streaming (cid:129) Whenauserrequestforavideoarrivesthenthisrequest
using Hadoop Myoungjin et al. [39] discussed the mul- issenttotheindexnode.Astherequestreachestheindex
timedia applications and services that require very high node, the load balancing mechanisms are executed and
processing and computing technologies. There are certain askedfortherequirecontentholdernode.
issueslikeloadbalancing,faulttolerance,andtaskmanage- (cid:129) Videosaredividedintomultiplesegmentsandstoredon
mentthatcouldaffecttheperformanceofmediaapplications. different content nodes. Content nodes are checked to
To overcome these issues and enhance the performance of findtherequiredcontentscontainingnodes.
123
46 JReliableIntellEnviron(2016)2:37–49
(cid:129) Atclientsidetheindexnodeinquiresthestreamingnodes Table1 Inthissection,wepresentthecriticalanalysisofthepresented
forbandwidthrequirementandforwardsthisbandwidth state-of-the-artrelatedtoresourceallocationandtaskmanagementand
alsopresentthecorrespondingopenresearchissues
reporttorequiredcontentcontainingnodes.
(cid:129) Inorder tofind theload balancing capacities streaming S.no. Issue Description
nodes are examined for their hardware resources. After
1 QoE Itistheextenttowhichuser
doing this results are handed over to the load balanc-
acceptstheapplication
ingmodelforcontentcapturing,analysis,separation,and
2 QoS Itistheefficiencyofthe
streaming. systemtoperformuser
(cid:129) Indexnodeasksforsettingthequalityleveltothestream- tasks
ingnodeandgivesinformationabouttherequiredcon- 3 Costoptimization Itistheprocess,inwhich
tent.So,thestreamingnodecandownloadtherequired cloudproviderminimizes
theuser’scosttorentthe
video.
cloudplatform
(cid:129) The downloaded content is analysed and separated
4 Serviceresponsetime Itistime,inwhichuser
accordingtotheH.264/SVCcodingscheme.Finally,the requestarriveatcloud
addressofselectedstreamingnodeissenttoindexnode manager,processedbythe
anditforwardsthataddresstoclientside. cloudandthensendresult
backtouser
5 Taskmanagement Itistheschedulingofuser’s
tasksinthecloudplatform
6 Resourceallocation Itistheallocationof
QoS/QoEmappingandadjustmentmodelinthecloud-based
resourcestoincoming
multimedia infrastructure Cloud-based multimedia system user’srequests
interactswithcurrentIP-basednetworkwhereitfacesQoS
and multi-cast service support challenges [41]. The cloud
Their proposed scheme contains certain features as fol-
providers ensure the QoS in the system. They neglect user
lows: it is predictive, it records the history of bandwidth
perspective which is an important factor in the design of
utilization for each video channel, and estimates the com-
cloud-based multimedia systems. QoE describes that how
ingrequirementsofclients.Moreover,achannelinterleaving
cloud-basedmultimediasystemwouldenhancetheusability
schemeisusedforvideocontents.Thisschemeprovideshigh
of the users. QoE is categorized into two further subcat-
qualityvideostotheclientsbyensuringthedemandedvideo
egories named objective and subjective. When system is
qualityandguaranteesthelessutilizationofnetworkband-
monitored for technical parameters, i.e. throughput, delay,
width.
andpacketloss,itiscalledobjectiveapproachandwhenit
ThefollowingTable1presentstheimportantperformance
monitorsthesystemonuseropinion,thenitiscalledsubjec-
metricsusedtoevaluatetheperformanceofMCCandtheir
tiveapproach.
correspondingshortdescription.
Inmultimediasystems,userperspectivehasalwaysbeen
lacking.Thus,Wei-Tingetal.[41]proposedaQoEmapping
and adjustment model that worked by translating the net- 4 Futuredirectionsandopenissues
workQoSparametersintotheuserQoEunderacloud-based
multimedia infrastructure. The proposed model consists of (1) Queuing model: Queuing model is an efficient mecha-
threemajorsectorsandthesimulationresultsshowedthatthe nism but there is a possibility of wastage of resources.
user’sQoEandnetworkQoSareconsistentwitheachother. Aseveryqueuehasaitsownseparatecomputingunit,
TheserviceprovidercouldusetheproposedQoEfunctionto it is not possible to keep the computing resource busy
monitortheusers’QoEperceptionandtorespondquicklyto allthetimeaswellasqueueswork-loaded.Hence,more
rectifyproblemsthatdegradedQoEinthemultimediacloud. accurate and efficient mechanisms are required which
Quality-assuredcloudbandwidthauto-scalingforvideo-on- minimizetheresourcewastageandimprovethesystem
demandapplications TheVideoonDemand(VoD)providers performance.
usedcloudcomputingbandwidthresourcestoguaranteethe (2) Priority services-based queuing model: Priority queu-
availabilityofVoDtotheclients.Authorsin[42]proposeda ingschemeisanefficientmechanismbutthereisahigh
predictive resourceauto-scaling systemwhichdynamically possibilityofincreasedserviceresponsetime.Asevery
booked the minimum bandwidth resources from multiple queueisprocessedaccordingtopredefinedpriorityand
data centres for the VoD provider to match its short-term ifanyurgenttaskneedstobeprocessedthenitwillnot
demandprojections.Theproposedarchitectureenhancesthe be processed until its priority arrives. There is a need
performance of video streaming using cloud-based auto- to enhance this priority queuing scheme for the same
scalingbandwidthmechanism. reason.
123