IEEE Wireless Communications June 201314 1536-1284/13/$25.00 2013 IEEE
MO B I L E CLOUD CO M P U T I N G
INTRODUCTIONWith the explosive growth of mobile devices inrecent years, there is a shift of user preferencesfrom traditional cell phones and laptops tosmartphones and tablets. Advances in the porta-bility and capability of mobile devices, togetherwith widespread third/fourth generation(3G/4G) Long Term Evolution (LTE) networksand WiFi access, have brought rich mobileapplication experiences to end users. Undoubt-edly, the demand for ubiquitous access to awealth of media content and services will con-tinue to increase, as indicated in a report by
Cisco: traffic from mobile devices is anticipatedto account for 60 percent of the total global IPtraffic by 2016.
However, the resource-constrained nature ofmobile devices, especially the limited battery life,has been a stumbling block to further improve-ments of mobile applications and services.According to the 2012 U.S. Wireless Smart-phone Customer Satisfaction Study, battery lifeis the least satisfying aspect of smart phones, andis one of the few attributes that have a declinedsatisfaction score among consumers, rating anaverage score of 6.7 out of 10 in 2012, downfrom 6.9 in 2011. While new smart phones withfaster CPUs, larger storage, and bigger screensare launched every day, and the bandwidth ofwireless networks has increased from kilobits persecond to megabits per second in just a fewyears, the development of batteries has laggedfar behind the development of other compo-nents (e.g., processors, storage, networks, anddisplays) in mobile devices. In fact, faster CPUsand larger displays consume more battery ener-gy: among users of 4G-enabled smart phones,satisfaction with battery performance is rated as6.1, which is much lower than the 6.7 battery sat-isfaction score of 3G users. Moreover, despitethe fast development of hardware technology, itis still difficult to support computation-intensiveapplications (e.g., image processing, augmentedreality) on mobile devices, hindering developersfrom bringing richer experiences and complexapplications to mobile users.
The paradigm of mobile cloud computing(MCC) is introduced to resolve the conflictsmentioned above, in which the cloud serves asa powerful complement to resource-constrainedmobile devices. MCC is a model for elastic aug-mentation of mobile device capabilities viaubiquitous wireless access to cloud storage andcomputing resources, with context-awaredynamic adaption to changes in the operatingenvironment. Rather than conducting all com-
FANGMING LIU, PENG SHU, HAI JIN, LINJIE DING, AND JIE YU, HUAZHONG UNIVERSITY OFSCIENCE AND TECHNOLOGY
DI NIU, UNIVERSITY OF ALBERTABO LI, THE HONG KONG UNIVERSITY OF SCIENCE AND TECHNOLOGY
ABSTRACTMobile cloud computing, with its promise to
meet the urgent need for richer applications andservices of resource-constrained mobile devices,is emerging as a new computing paradigm andhas recently attracted significant attention. How-ever, there is no clear definition and no welldefined scope for mobile cloud computing dueto commercial hype, and diverse ways of com-bining cloud computing and mobile applications.This article makes the first attempt to present asurvey of mobile cloud computing from the per-spective of its intended usages. Specifically, weintroduce three common mobile cloud architec-tures and classify comprehensive existing workinto two fundamental categories: computationoffloading and capability extending. Consideringthe energy bottleneck and user context of mobiledevices, we discuss the research challenges andopportunities of introducing cloud computing toassist mobile devices, including energy-efficientinteractions, virtual machine migration over-head, privacy, and security. Moreover, wedemonstrate three real-world applicationsenabled by mobile cloud computing, in order tostimulate further discussion and development ofthis emerging field.
GEARING RESOURCE-POOR MOBILE DEVICES WITHPOWERFUL CLOUDS: ARCHITECTURES,
CHALLENGES, AND APPLICATIONS
LIU LAYOUT_Layout 1 6/21/13 11:58 AM Page 14
IEEE Wireless Communications June 2013 15
putational and data operations locally, MCCtakes advantage of the abundant resources incloud platforms to gather, store, and processdata for mobile devices. Many popular mobileapplications have actually employed cloud com-puting to provide enhanced services. For exam-ple, Apples iCloud enables users to store andsynchronize their data (photos, videos, contacts,etc.) in the cloud. Mobile web browsers likeAmazon Silk (Dolphin, Opera, etc.) exploit thecomputation and network resources in AmazonElastic Compute Cloud (EC2) to compress webpages, making it faster and cheaper to down-load these pages to mobile devices. More inno-vative cloud-based mobile applications likehealthcare monitoring and massively multiplay-er online (MMO) mobile games are also underdevelopment.
Cloud computing and mobile applicationshave been the leading technology trends inrecent years. It is not surprising that MCC,combining the two, would create a thrillingfield and attract attention from both industryand academia. But what is MCC? Does it reallybring a novel and useful computing paradigm,or is it just commercial hype? In what ways canmobile devices benefit from the cloud? Andwhat are the biggest challenges of using cloudcomputing to augment and enhance mobileapplications?
To answer the above questions, the rest ofthis article is organized as follows. First, we startwith an overview of the existing architectures ofMCC. Second, we analyze how to exploit cloudresources to benefit mobile devices, classifyingexisting usage models into two categories: com-putation offloading and capability extending.Third, we unveil the challenges in gearing mobiledevices with the cloud. Finally, we highlightthree representative killer applications driven byMCC before concluding this article.
ARCHITECTURES AND VISIONSWe first provide an overview of the architecturesunderlying MCC, which define the way in whichmobile devices are connected to and interactwith the cloud. Specifically, we discuss the tradi-tional centralized cloud, as well as the recentlyproposed cloudlet and peer-based ad hoc mobilecloud scenarios, followed by our visions forfuture integrated mobile cloud architectures.
CENTRALIZED CLOUDWe refer to the architecture where mobiledevices obtain services from a traditional datacenter as a centralized cloud . In this archi-tecture, the cloud resource is placed in aremote centralized cloud infrastructure, asil lustrated in Fig. 1. To access data centerresources, mobile devices usually access thebackbone network via WiFi access points (APs)or 3G/4G cellular networks. The cloud hereacts as an agent between the original contentproviders and mobile devices. Rather than run-ning applications locally and directly request-ing data from content providers, a mobiledevice can offload parts of their workload tothe cloud, taking advantage of the abundantcloud resources to help gather, store, and pro-cess data for the mobile device. Recently, manysolutions based on this architecture haveemerged to enhance mobile experiences, onwhich we further elaborate in Table 2.
CLOUDLETThe access to the centralized cloud incurs longlatency due to wide area network (WAN) delays,as illustrated in Fig. 1. However, real-timemobile applications (e.g., online games, speechrecognition, Facetime) have strict requirementson response time. Users of these applicationsmay have unsatisfactory experiences due to
Figure 1. Centralized cloud and cloudlet architecture.
Rather than runningapplications locally anddirectly requesting datafrom content providers,
a mobile device canoffload parts of their
workload to the cloud,taking advantage of the
abundant cloudresources to help gather,store, and process datafor the mobile device.
LIU LAYOUT_Layout 1 6/21/13 11:58 AM Page 15
delayed communication between the mobiledevice and cloud.
Cloudlet  has emerged as a promisingarchitecture to reduce such communicationdelay. A cloudlet is a resource-rich server thathas Internet access and is well connected tomobile devices via a high-speed local area net-work (LAN), as highlighted in Fig. 1. Smoothinteraction can easily be achieved because of thelow latency in one-hop LANs. This paradigm issimilar to that of WiFi hotspots nowadays, wherea mobile device gains high-speed Internet accesswhenever a WiFi hotspot is nearby. However,many practical issues remain to be addressed,including the sparseness of cloudlet deployment,the business model, and cloudlet reliability.Although there is a long way to go before thecloudlet can be deployed, we believe that it hasinspired a new MCC paradigm.
AD HOC MOBILE CLOUDThe above architectures assume persistent con-nectivity to either the cloud or cloudlet, which isnot always available or affordable (e.g., when themobile device is roaming). A third architectureregards a crowd of mobile devices as an ad hocmobile cloud, where the neighboring mobiledevices are pooled together for resource sharing.As shown in Fig. 2, in this paradigm, a task froma mobile device can be either processed in a dis-tributed and collaborative fashion on all themobile devices or handled by a particular mobiledevice that acts as a server. The feasibility of thisarchitecture has been demonstrated in VirtualCloud Provider , a peer-based MapReduceframework for mobile devices. It organizes near-by mobile devices pursuing the same task into avirtual cloud to distribute computation tasks.
VISIONS FOR THE FUTURE ARCHITECTUREFirst, we envision that future MCC will leveragea combination of different architectures, withtraditional data-center-based solutions on oneend of the spectrum and the ad hoc mobile cloudon the other end. An ad hoc mobile cloud is abackup solution when a group of users in collec-tion are unable to access WiFi or cellular net-
works (e.g., passengers in the subway). Evenwith good network access, the ad hoc mobilecloud may be desirable for highly parallel tasksdue to its crowd-based nature. Cloudlets areplaced between mobile users and data centers,and are preferable for highly interactive anddelay-sensitive applications (e.g., computervision, augmented reality). Cloudlets can bedeployed along with wireless APs and be chargedfor in a pay-as-you-go fashion similar to tradi-tional cloud computing.
Second, the mobile cloud framework shouldsupport strong fault tolerance and transparentmigration across different architectures toachieve seamless integration. A device may loseits connection to the centralized cloud during aservice due to interrupted network connectivity.In an ad hoc mobile cloud, peers may unexpect-edly quit the ad hoc hot zone. Under such situa-tions, the system should recover from the faultas quickly as possible, and the current computingstate should be suspended, migrated to theaccessible cloud providers, and resumed; other-wise, the related devices will have to spend extratime on native execution. Such seamless integra-tion and switching between different architec-tures require an open interface standard forheterogeneous cloud platforms and mobiledevices.
Last but not least, mobile devices should playthe role of not only independent consumers, butalso interdependent contributors. Besides beingresource providers in the ad hoc mobile cloud,mobile devices can be service participants. Vari-ous sensors (e.g., cameras, GPS, microphones)have made the mobile device a natural collectorof context information. For example, mobiledevices in vehicles can report their local trafficconditions to the centralized cloud. With trafficinformation collected from all over the city, thecloud can predict traffic jams or suggest routesfor drivers. Moreover, a mobile device can serveas a resource relay; for example, nodes with sta-ble Internet access in an ad hoc group couldshare the network to other mobile devices withpoor connections, keeping the ad hoc mobilecloud connected to the Internet.
The seamless integration of heterogeneouscloud platforms and mobile devices, togetherwith the proper utilization of each mobile inhab-itant, will finally make mobile cloud computing astrong ecosystem. With the mobility and contextawareness of mobile devices, such an ecosystemcould better connect the cyber world with the phys-ical world, eventually forming a foundation ofhuman-centric computing, exemplified by aug-mented reality, healthcare, and personal assis-tance, as demonstrated later.
INTENDED USESBased on the above representative architectures,we proceed to discuss how mobile devices canleverage cloud computing for various uses. Con-sidering the bottleneck of battery life and thedemand for richer services in mobile devices,existing work can be categorized into two class-es: computation offloading and capability extend-ing, along with a detailed taxonomy summarizedin Tables 1 and 2, respectively.
IEEE Wireless Communications June 201316
Figure 2. Ad hoc mobile cloud and its two working modes.
Ad hoc mobile cloud
Requesting a particular service
LIU LAYOUT_Layout 1 6/21/13 11:58 AM Page 16
IEEE Wireless Communications June 2013 17
COMPUTATION OFFLOADINGTo overcome resource constraints on mobiledevices, a general idea is to offload parts ofresource-intensive tasks to the cloud. Since exe-cution in the cloud is considerably faster thanthat on mobile devices, it is worth shipping codeand data to the cloud and back to prolong thebattery life and speed up the application. Thisoffloading framework is illustrated in Fig. 3.Despite diverse technologies to realize the run-time environment in the cloud (Table 1), themajor differences between offloading techniqueslie in the offloading unit and partitioning strate-gies.
Offloading Existing Applications The first class ofapproaches attempts to extract offloading-friend-ly parts of codes from existing applications.MAUI  proposes a model to enable fine-grained code offloading to the cloud. To lever-age MAUI, developers need to manuallyannotate which methods could be offloaded. Aruntime profiler then predicts the energy usageof the method invocation according to currentstatus information, and decides whether amethod should be executed natively or remotelyin order to minimize energy consumption.
Although the fine-grained method-leveloffloading strategy used in MAUI may maximizeenergy savings in mobile devices, it takes a con-siderable amount of time for programmers toannotate offloadable methods. To avoid manual
annotation, CloneCloud  automaticallydecides what could be offloaded by utilizing astatic code analyzer to mark possible migrate/merge points in the program bytecode. It thenutilizes dynamic profiling to determine the opti-mal partitioning under different computationand network environments.
Whereas MAUI and CloneCloud can onlyoffload one method/thread at a time and aretroubled by locking issues, COMET  over-comes these limitations by relying on distributedshared memory (DSM) systems and virtualmachine (VM) synchronization techniques toenable multithreaded offloading. A field-levelgranularity is used to manage memory consisten-cy in order to reduce the frequency of requiredcommunication between the mobile device andcloud.
ThinkAir  also tries to extend MAUI andCloneCloud by further exploiting the auto-scal-ing feature offered by the cloud. For example,when the resource depletion (e.g., out of memo-ry) of a clone VM is detected, instead of propa-gating the exception back, the cloud will allocatemore VMs for the task. Moreover, if the clouddetects that the computation is parallelizable, itwill automatically launch multiple VMs to exe-cute the job in parallel.
New Development Models While the above meth-ods all focus on offloading existing applications,offloading can also be taken into account duringthe initial development stage of new applica-
Table 1. Comparison of computation offloading models.
Approaches Blocking Optimization Offloading unit Concurrency ConceptC
gMAUI Battery Method X
CloneCloud Battery and performance Thread X
COMET X Battery and performance Multithreads X
ThinkAir Battery Method
mCloud X Performance Component Component-basedapplication
Weblet X Battery Component
Table 2. Comparison of approaches in capability extension.
Approaches Feature Improved functions NetworktrafficMajor resourceprovided by cloud
Silk Web browser Loading speed, data volume Medium Computation
Voice Search/Siri Natural language UI Speech recognition, semantic analysis Medium Computation
iCloud/Dropbox Extended storage Storage capacity, file sharing High Storage
CloudTorrent File downloading Download speed High Network
Reflex Social interaction Spontaneous and global interaction Low Network
MobiCloud Network optimization Informed and secure routing/communication Low Computation
LIU LAYOUT_Layout 1 6/21/13 11:58 AM Page 17
IEEE Wireless Communications June 201318
tions. Mobile applications usually consist of vari-ous components serving different high-levelfunctionalities, many of which are reusable (e.g.,image processing, face recognition). In a newmodel, offloadable components are predefinedin the development phase, and are shippedbetween mobile devices and the cloud in an on-demand manner. For example, MCloud  intro-duces a composition approach that treats anapplication as a combination of reusable compo-nents. Component developers can publish to themarket their developed components, based onwhich other developers can easily build a newapplication.
Weblet  adopts a similar design, in whichan application is composed of weblets. A webletis a platform-independent entity that presents astateless and HTTP-based web service interface.The framework dynamically decides whether aweblet should run on the mobile device or in thecloud, based on attributes such as power con-sumption, monetary cost, and performance. Fur-thermore, a weblet in the cloud can replicateitself in the forms of pool and shadow. Pool han-dles multiple tasks in parallel, while shadow pro-vides fault tolerance and latency control.
CAPABILITY EXTENDINGBesides saving energy, cloud computing alsoenables enhanced mobile experiences that werepreviously impossible on resource-constrainedmobile devices. Many commercial mobile appli-cations use the cloud to bring about rich fea-tures, as illustrated in Table 2. They usuallyemploy a client-server framework that consists oftwo parts, which run on the mobile device andthe cloud, respectively. Essentially, cloud com-puting helps extend the capabilities of mobiledevices in three aspects: computation, storage,and networking.
Computational Capability Many applications nowa-days support speech recognition (e.g., GoogleVoice Search, Apple Siri). The acoustic modelsfor recognition and high-quality speech synthesismust be trained with millions of voice samples inthousands of dialects. This computation-inten-sive task is infeasible on a mobile device andshould be delegated to the cloud. Only there
can you reach the scale, the enormous volume ofinteractions required to create a speech systemcapable of rivaling human understanding, saidLarry Heck, Microsofts chief scientist in theSpeech group and former vice president ofNuance.
Storage Another class of popular mobile cloudapplications aims to extend storage, in which userdata are stored in the cloud and synchronized topervasive devices. Dropbox, a popular cloud stor-age service, uses Amazons S3 storage system asits back-end, and secures user data with AES-256encryption. To minimize synchronization cost, itutilizes binary-delta encoding to upload thechanges made to a file rather than the entire file.Apples iCloud is another storage service thatenables users to store or access data like applica-tions, music, and contacts in the cloud throughmultiple devices, including iOS-based devices andPCs. It uses both Microsoft Azure and AmazonWeb Services as its service hosts. Although thereis steady growth in mobile storage capacity, theever increasing appetite of users for high-resolu-tion videos and images promises the increasingpopularity of cloud storage.
Networking The popularity of mobile devicesplaces higher requirements on network availabil-ity and stability. Unlike a mobile device that haslimited network connectivity, a cloud provider istypically connected to multiple carriers andInternet service providers (ISPs) with high-speedlinks, and can access Internet resources easily.Therefore, the cloud also promises to be a pow-erful agent in network connectivity improve-ment.
For instance, CloudTorrent  employs thecloud as an agent between content providers andmobile devices, which builds a virtual link formobile devices to access Internet resources. Itlets the cloud run BitTorrent file sharing appli-cations and send downloaded files to mobiledevices to achieve fast and energy-efficient filedownloading. Reflex  proposes a flexible andreusable system framework for social networkapplication development, where worldwide userscould spontaneously be connected together forconferencing and gaming via the cloud. It
Figure 3. The abstract procedure of computation offloading.
Virtual machine monitorHardware
Remote-able part(method, thread, component)
Unlike a mobile devicethat has limited networkconnectivity, a cloudprovider is typically con-nected to multi-carriersand ISPs with high-speed links, and canaccess Internet resourceseasily. Therefore, thecloud also promises tobe a powerful agent innetwork connectivityimprovement.
LIU LAYOUT_Layout 1 6/21/13 11:58 AM Page 18
IEEE Wireless Communications June 2013 19
exploits well connected data centers across dif-ferent regions of the same cloud provider to sup-port real-time interactive sessions with lowlatency. MobiCloud  builds up a secure ser-vice-oriented mobile cloud framework to assistcommunications in mobile ad hoc networks(MANETs). It employs the cloud to create a vir-tualized environment where physical devices aremirrored to so-called extended semi-shadowimages (ESSIs), creating a virtualized MANETrouting layer that can guide routing among realmobile devices.
CHALLENGES AND OPPORTUNITIESWe further identify four critical challenges andopportunities in MCC, with respect to thestochastic characteristic of wireless networks, vir-tual machine (VM) migration overhead, privacy,and security.
ENERGY-EFFICIENT INTERACTIONSWireless networks are stochastic in nature: notonly the availability and network capacity (e.g.,bandwidth, signal strength) of APs vary fromplace to place, but the downlink and uplinkbandwidth also fluctuates due to weather, build-ing shields, mobility, flash crowds, and so on, asshown in Fig. 1. This stochastic characteristicmay incur unpredictable energy consumption incommunications between mobile devices and thecloud: measurement studies  show that theenergy consumption for transmitting a fixedamount of data is inversely proportional to theavailable bandwidth. This implies that transmit-ting data in bad connectivity could consume con-siderably more energy than doing so in goodconnectivity.
Energy saving from computation offloading isnot guaranteed on mobile devices if the evokeddata transfers via wireless networks consume anunpredictable amount of energy. Offloading cansave energy only if heavy computation is neededand a relatively small amount of data has to betransferred. Energy efficiency can be substantial-ly improved if the cloud stores the data requiredfor computation, reducing data transmissionoverhead. Admission control and bandwidthallocation mechanisms in cellular base stationsand APs may guarantee network connectivity toa certain extent, but cannot eliminate thestochastic nature of wireless links. An alternative
approach is to dynamically adjust applicationpartitioning between the cloud and mobiledevices according to network conditions [4, 5],although it is challenging to quickly and accu-rately estimate the network connectivity with lowoverhead.
Energy-efficient communications are alsocritical when exploiting the cloud to extend thecapabilities of mobile devices. Frequent trans-missions in bad connectivity will overly consumeenergy, making the extended capabilitiesunattractive, as battery life is always the top con-cern of mobile users. The latest practical solu-tion, called eTime , is to adaptively seize thetiming opportunity when network connectivity isgood to prefetch frequently used data whiledeferring delay-tolerant data. However, thisapproach is mainly suitable for prefetch-friendlyor delay-tolerant applications, such as social net-working services (SNSs) and cloud storage.
VM MIGRATION OVERHEADElastic resource scaling is one of the most impor-tant advantages of cloud computing. Enabled bythe virtualization technology, cloud resourcescan be provisioned as needed to complementmobile applications ; that is, a VM instancecan be launched when user demands arise, andshut down after the task terminates. However,the dynamic VM provisioning and time-varyingresource demand of each VM instance may leadto underutilization on the underlying physicalservers. To enhance utilization and energy effi-ciency, cloud operators need to periodicallymigrate and consolidate VM instances acrossphysical servers and even set some servers topower saving mode if they are not used.
The above VM migration mechanisms mayimpact the performance of mobile cloud applica-tions: not only does the migration itself incursuspended execution time, but the resource com-petition between different VMs hosted on thesame physical server may also delay task comple-tion. A possible remedy is that since the applica-tions are processed at both the cloud and mobilesides, migration can be conducted when the cur-rent task is executed on the mobile side. More-over, we can carefully group the tasks thatrequire different types of resources on a server.For example, a CPU-intensive image processingtask and a network-intensive video streamingtask can be placed on the same physical server
Figure 4. SixthSense1 (left) and Project Glass2 (right).
1 The thrilling potentialof SixthSense technology:http://on.ted.com/bDed
2 Project Glass: Oneday: http://www.youtube.com/watch?v=9c6W4CCU9M4
LIU LAYOUT_Layout 1 6/21/13 11:58 AM Page 19
IEEE Wireless Communications June 201320
to alleviate resource contention, eventuallyimproving front-end quality of experience ofmobile cloud applications.
On the other hand, VM migration mecha-nisms can actually be leveraged to improve taskperformance of mobile cloud applications; forexample, a VM instance can be migrated from abusy server to a lightly loaded one to speed upexecution. VM instances can also be adaptivelymigrated across geographically distributed datacenters, as the mobile user moves, to reduce theinteraction latency between the mobile deviceand cloud platform. The trade-off betweenmigration overhead and performance in mobilecloud computing should be carefully tuned andbalanced under different scenarios.
PRIVACYSince mobile devices are usually personal items,privacy must be considered when leveraging thecloud to store and process their confidential dataremotely. Many mobile users are concerned withadvertisement pushes, through which their pri-vate information like hobbies and locations maybe unconsciously collected and illegally spread.
Huang et al. propose a secure data processingframework for mobile cloud computing , inwhich critical data are protected by the uniqueencryption key generated from the users trustedauthority and stored in an area isolated from thepublic domain. Even when storage is breached inthe cloud, unauthorized parties including thecloud vendor cannot obtain the private data. Butsuch encryption cannot handle scenarios inwhich the cloud needs to operate on data (e.g.,spelling checks).
Another particular privacy issue for mobileusers is the leakage of personal location infor-mation in location-based services. To address theproblem, a method called location cloakingmakes user location data slightly imprecisebefore submitting them to the cloud . Butthe imprecise data sometimes cannot providerelevant or satisfactory results to users in certainapplications. Therefore, location cloaking shouldbe adaptively tuned to balance the trade-offbetween privacy and result accuracy .
SECURITYThere are several aspects of mobile cloud securi-ty, including antivirus, authentication, data pro-tection, and digital rights management. Securityvulnerability can cause serious problems, includ-ing property damage, cloud vendor economicloss, and user distrust. There are many instancesof malware attempting to steal personal informa-tion or intercept mobile transactions. Sincemobile devices are resource-constrained, locallyexecuted antivirus software can hardly protectthem from threats efficiently. A current solutionis to offload the threat detection functionality tothe cloud. Nevertheless, since a pure cloudantivirus relies on cloud resources, it is difficultto deal with malware that can block the devicesInternet connection.
Besides, authentication is critical for access tosensitive information such as bank accounts andconfidential files. With constrained text input onmobile devices, users tend to use simple pass-words, making mobile applications more vulner-
able to authentication threats. To solve thisissue,  builds up an authorization platformwhere users are identified by their habits (e.g.,calling patterns, location information, and webaccess). The platform routinely records userbehavior information. When a server receives anauthorization request, it redirects the request toan authorization engine, which uses the aggre-gated behavior information and an authorizationpolicy to decide whether to accept the request ornot.
A new class of mobile applications, augmentedreality (AR), has started to draw users atten-tion. Wearable mobile devices, like gesturalinterface SixthSense and Googles head-mounteddisplay Project Glass, aim to blur the boundarybetween the cyber world and real world. Forexample, as shown in Fig. 4, SixthSense can pro-ject augmented live news on a real-world news-paper; Google Glass can overlay wearers visionwith map directions, calendar reminders, textmessages, and so on. Augmented reality is alsoincorporated into mobile games, where virtualobjects are projected into the real world so thatusers can interact with them. Nevertheless, algo-rithms in augmented reality are mostly resource-and computation-intensive, posing challenges toresource-poor mobile devices.
These applications can integrate the power ofthe cloud to handle complex processing of aug-mented reality tasks. Specifically, data streams ofthe sensors on a mobile device can be directedto the cloud for processing, and the processeddata streams are then redirected back to thedevice. It should be noted that AR applicationsdemand low latency to provide a life-like experi-ence. In this sense, apart from exploiting cloudresources, a mobile device can also offload dataprocessing to a nearby cloudlet or ad hoc mobilecloud as elaborated earlier to avoid unpre-dictable multihop network latencies.
REMOTE HEALTHCAREWith the increasing popularity of mobile devices,all industries are changing their business andproducts to adapt to mobile technology. Health-care  is one of the leading areas where smartphones, tablets, and sensors are deployed on alarge scale. Through remote communication,doctors and nurses can get a real-time picture ofpatients conditions and perform the correspond-ing treatments. For example, Rehabcare GroupInc., which runs hospitals and medical facilitiesthroughout the United States, has built cloud-based apps on SalesForece.com to improvehealthcare experiences, such as paperless patientpreadmission and screening, and remote post-operative monitoring.
A body area network is another example ofapplying mobile technology in healthcare. Pro-jects like CodeBlue and CareNet deploy smallwearable sensors on elderly people, monitoringtheir physical conditions (e.g., temperature,heart rate). These systems need to employ cloudcomputing to simulate and analyze the massive
Since mobile devices areusually personal items,privacy must be consid-ered when leveragingthe cloud to store andprocess their confidentialdata remotely. Manymobile users are con-cerned with advertise-ment pushes, throughwhich their private infor-mation like hobbies andlocations may be uncon-sciously collected andillegally spread.
LIU LAYOUT_Layout 1 6/21/13 11:58 AM Page 20
IEEE Wireless Communications June 2013 21
data collected. When emergencies like myocar-dial infarction and falling down are detected,hospitals aree informed immediately to providethe necessary help.
CONCLUSIONSIn this article, we describe the state of the art ofmobile cloud computing, from the brewingoffloading technology to commercialized mobilecloud applications. We unveil the major challengesin mobile cloud computing and discuss potentialsolutions. Although mobile devices geared towardcloud computing will undoubtedly change technol-ogy trends as well as our daily lives, many practicalproblems remain to be resolved to structure a full-fledged mobile cloud system.
ACKNOWLEDGMENTThe corresponding author is Hai Jin. Theresearch was supported by a grant from theNational Natural Science Foundation of China(NSFC) under grant No. 61133006.
REFERENCES Z. Liu et al., Socialize Spontaneously with Mobile
Applications, Proc. IEEE INFOCOM, 2012. M. Satyanarayanan et al., The Case for VM-based
Cloudlets in Mobile Computing, IEEE Pervasive Com-puting, vol. 8, no. 4, 2009, pp. 1423.
 G. Huerta-Canepa et al., A Virtual Cloud ComputingProvider for Mobile Devices, Proc. ACM Wksp.MobileCloud Computing & Services: Social Networks andBeyond, 2010.
 E. Cuervo et al., Maui: Making Smartphones LastLonger with Code Offload, Proc. ACM MobiSys, 2010.
 B. Chun et al., Clonecloud: Elastic Execution BetweenMobile Device and Cloud, Proc. ACM EuroSys, 2011.
 M.S. Gordon et al., COMET: Code Offload by Migrat-ing Execution Transparently, Proc. USENIX OSDI, 2012.
 S. Kosta et al., Thinkair: Dynamic Resource Allocationand Parallel Execution in the Cloud for Mobile CodeOffloading, Proc. IEEE INFOCOM, 2012.
 V. March et al., mCloud: Towards a New Paradigm ofRich Mobile Applications, Proc. MobiWis, 2011.
 X. Zhang et al., Towards an Elastic Application Modelfor Augmenting the Computing Capabilities of MobileDevices with Cloud Computing, Springer Mobile Net-works & Applications, vol. 16, no. 3, 2011, pp. 27084.
 I. Kelenyi et al., CloudTorrent Energy-Efficient Bit-Torrent Content Sharing for Mobile Devices via CloudServices, Proc. IEEE CCNC, 2010.
 D. Huang et al., MobiCloud: Building Secure CloudFramework for Mobile Computing and Communica-tion, Proc. IEEE SOSE, 2010.
 N. Balasubramanian et al., Energy Consumption inMobile Phones: a Measurement Study and Implicationsfor Network Applications, Proc. ACM IMC, 2009.
 P. Shu et al., eTime: Energy-Efficient Transmissionbetween Cloud and Mobile Devices, Proc. IEEE INFO-COM, 2013.
 D. Huang et al., Secure Data Processing Frameworkfor Mobile Cloud Computing, Proc. IEEE INFOCOMWksp. 2011.
 R. Cheng et al., Preserving User Location Privacy inMobile Data Management Infrastructures, Proc. Wksp.Privacy Enhancing Technologies, 2006.
 R. Chow et al., Authentication in the Clouds: aFramework and Its Application to Mobile Users, Proc.ACM Wksp. Cloud Computing Security Wksp., 2010.
 M. Chen et al., A 2G-RFID Based E-healthcare Sys-tem, IEEE Wireless Commun., vol. 17, no. 1, 2010, pp.3743.
BIOGRAPHIESFANGMING LIU [M] (email@example.com) is currently anassociate professor in the School of Computer Science andTechnology, Huazhong University of Science and Technolo-gy, Wuhan, China; and he is the CHUTIAN Scholar of HubeiProvince, China. Since 2012 he has also been invited as amember of StarTrack Visiting Young Faculty at MicrosoftResearch Asia (MSRA), Beijing. He received his B.Engr.degree in 2005 from Department of Computer Science andTechnology, Tsinghua University, Beijing, China; and hisPh.D. degree in computer science and engineering fromthe Hong Kong University of Science and Technology in2011. From 2009 to 2010, he was a visiting scholar at theDepartment of Electrical and Computer Engineering, Uni-versity of Toronto, Canada. He was the recipient of twoBest Paper Awards from IEEE GLOBECOM 2011 and IEEECloudCom 2012, respectively. His research interests includecloud computing and data center networking, mobilecloud, green computing and communications, software-defined networking and virtualization technology, large-scale Internet content distribution, and video streamingsystems. He is a member of ACM, as well as a member ofthe China Computer Federation (CCF) Internet TechnicalCommittee. He has been a Guest Editor for IEEE Network,and served as TPC for IEEE INFOCOM 2013 and 2014 andGLOBECOM 2012 and 2013.
PENG SHU is currently a Masters student in the School ofComputer Science and Technology, Huazhong University ofScience and Technology, Wuhan, China. His research inter-ests focus on cloud computing and wireless mobile appli-cations.
HAI JIN [SM] is a Cheung Kung Scholars Chair Professor ofcomputer science and engineering at the Huazhong Univer-sity of Science and Technology (HUST), China. He is nowdean of the School of Computer Science and Technology atHUST. He received his Ph.D. degree in computer engineer-ing from HUST in 1994. In 1996, he was awarded a Ger-man Academic Exchange Service fellowship to visit theTechnical University of Chemnitz in Germany. He worked atthe University of Hong Kong between 1998 and 2000, andas a visiting scholar at the University of Southern Californiabetween 1999 and 2000. He was awarded the ExcellentYouth Award from the National Science Foundation ofChina in 2001. He is the chief scientist of ChinaGrid, thelargest grid computing project in China, and chief scientistof the National 973 Basic Research Program Project of Vir-tualization Technology of Computing Systems. He is amember of the ACM. His research interests include com-puter architecture, virtualization technology, cluster com-puting and grid computing, peer-to-peer computing,network storage, and network security.
LINJIE DING is currently a research intern in the School ofComputer Science and Technology, Huazhong University of
Although mobile devicesgeared toward cloud
computing will undoubtedly change the
technology trends aswell as our daily lives,
many practical problemsremain to be resolved to
structure a full-fledgedmobile cloud system.
LIU LAYOUT_Layout 1 6/21/13 11:58 AM Page 21
IEEE Wireless Communications June 201322
Science and Technology, Wuhan, China. His research inter-ests focus on cloud computing and wireless mobile appli-cations.
JIE YU is currently a research intern in the School of Com-puter Science and Technology, Huazhong University of Sci-ence and Technology, Wuhan, China. Her research interestsfocus on cloud computing and wireless mobile applica-tions.
DI NIU is currently an assistant professor in the Departmentof Electrical and Computer Engineering, University of Alber-ta, Canada. He received his B.Engr. degree from theDepartment of Electronics and Communication Engineer-ing, Sun Yat-sen (Zhongshan) University, Guangzhou,Guangdong, China, in 2005, and his M.A.Sc. and Ph.D.degrees from the Department of Electrical and ComputerEngineering, University of Toronto, Ontario, Canada, in2009 and 2012, respectively. His research interests includecloud computing, large-scale data analytics, parallel anddistributed optimization, network economics, statisticaland machine learning for Internet applications, time seriesanalysis, stochastic modeling, multimedia transmission, andnetwork coding.
BO LI [F] is a professor in the Department of Computer Sci-ence and Engineering, Hong Kong University of Scienceand Technology. He received his B.Engr. degree in comput-er science from Tsinghua University and his Ph.D. degree inthe electrical and computer engineering from the Universi-ty of Massachusetts at Amherst. He was with IBM Net-working System, Research Triangle Park, North Carolina,between 1993 and 1996. He was an adjunct researcher atMicrosoft Research Asia (MSRA) (19992006). He was withMicrosoft Advanced Technology Center (ATC) in the sum-mers of 2007 and 2008. He has made original contribu-tions on Internet proxy placement, capacity provisioning inwireless networks, routing in WDM optical networks, andInternet video streaming. He is best known for a series ofworks on a system called Coolstreaming (Google entriesover 1,000,000 in 2008 and Google scholar citations over800), which attracted millions of downloads and was cred-ited as the first large-scale peer-to-peer live video stream-ing system in the world. His recent work on thepeer-assisted online hosting system FS2You (20072009)(Google entries 800,000 in 2009) has also attracted mil-lions of downloads worldwide.
LIU LAYOUT_Layout 1 6/21/13 11:58 AM Page 22
/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages false /GrayImageMinResolution 300 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages false /GrayImageDownsampleType /Average /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages true /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages false /MonoImageMinResolution 1200 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages false /MonoImageDownsampleType /Average /MonoImageResolution 1200 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier (CGATS TR 001) /PDFXOutputCondition () /PDFXRegistryName (http://www.color.org) /PDFXTrapped /False
/CreateJDFFile false /Description > /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ > /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles true /MarksOffset 6 /MarksWeight 0.250000 /MultimediaHandling /UseObjectSettings /Namespace [ (Adobe) (CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /UseName /PageMarksFile /RomanDefault /PreserveEditing true /UntaggedCMYKHandling /LeaveUntagged /UntaggedRGBHandling /LeaveUntagged /UseDocumentBleed false >> > ]>> setdistillerparams> setpagedevice