Skip navigation
Please use this identifier to cite or link to this item: http://arks.princeton.edu/ark:/88435/dsp018049g5084
Full metadata record
DC FieldValueLanguage
dc.contributor.advisorMartonosi, Margaret Roseen_US
dc.contributor.advisorPeh, Li-Shiuanen_US
dc.contributor.authorKoukoumidis, Emmanouilen_US
dc.contributor.otherElectrical Engineering Departmenten_US
dc.date.accessioned2012-03-29T18:04:27Z-
dc.date.available2012-03-29T18:04:27Z-
dc.date.issued2012en_US
dc.identifier.urihttp://arks.princeton.edu/ark:/88435/dsp018049g5084-
dc.description.abstractMobile devices, such as smartphones and personal media players, have recently significantly increased in popularity thanks to the rich set of mobile cloud services that they allow users to access. Networked vehicular computing devices are also expected to be commonplace in the near future, as they will enable a wide range of driver assistance services. The ubiquitous penetration of mobile services, however, has been thwarted by their poor user experience; access to mobile cloud services typically occurs over slow and costly long-range cellular communications. This thesis focuses on improving the user experience of mobile services by reducing the need for costly long-range cellular communications. To achieve this, the thesis proposes to host more service functionality on mobile devices themselves. In this way, mobile devices are often able to serve requests either locally or by contacting neighbor devices over short-range communications. Two novel mobile service architectures are proposed for the two different types of mobile services: traditional non-geo-locality services and emerging geo-locality services. A service is termed to have the geo-locality property when its data are both generated (sensed or input) and consumed locally, i.e., within a specific geographic region. In other words, for services that have the geo-locality property, only mobile devices within a specific geographic region <italic>R</italic> can generate the necessary service data and only devices within the very same region <italic>R</italic> are interested in consuming it. For non-geo-locality services, the data is generated either by cloud servers or by users regardless of their location. Data generation and/or consumption are also typically a function of the users' personal interests and not of their geographic location. For traditional non-geo-locality services, this thesis proposes the Pocket Cloudlets architecture. The Pocket Cloudlets architecture is a mobile device-resident caching scheme that serves cloud service requests locally on the device, when possible, significantly reducing the need for slow and costly long-range communications. The Pocket Cloudlets architecture leverages both personal user and collaboratively-generated community access patterns to selectively replicate parts of the cloud service locally on the mobile device. Pocket Cloudlets are also adaptively updated by detecting emerging popular service data items and prefetching them on the mobile device. Our analysis shows that the proposed Pocket Cloudlets architecture can effectively augment several traditional cloud services, like mobile web search. PocketSearch, our prototyped mobile search pocket cloudlet, reduces the average service access time by a factor of 2.7&times; and the required communication bandwidth by 66&#37;. For emerging geo-locality services, the thesis presents the Region-Resident Services (RegReS) middleware. RegReS allows a rich set of emerging geo-locality services to be fully supported on confederations of mobile devices. Mobile devices collaborate to provide a geo-locality service within a specified region and over a specified service lifetime by utilizing only short-range ad-hoc communications. In this way, RegReS completely eliminates the need for long-range cellular communications. Although mobile devices are becoming increasingly powerful, their resources are constrained and should be used judiciously. RegReS enables the efficient provision of geo-locality services by allowing services to specify their target service carrier density. Only as many service carriers as specified are subsequently maintained by RegReS. As opposed to previously proposed static schemes, RegReS employs a fully distributed, collaborative and adaptive estimation scheme to track the existing service carrier density and make decisions about the spawning of new carriers, when necessary. Thanks to collaboration and adaptation, RegReS can maintain the desired density with only 16&#37; mean absolute error across a wide range of configurations. To demonstrate the potential of collaborative mobile device-based computing platforms that are enabled by middleware like RegReS, the thesis presents a rich set of novel services that such platforms can enable. More specifically, the thesis focuses on the type of services that are typically most challenging and resource-intensive (e.g., CPU), camera-based services. We introduce five such services and prototype SignalGuru, a camera-based traffic signal schedule advisory service. SignalGuru leverages opportunistic sensing and collaboration across windshield-mounted smartphones and their cameras to provide drivers with information about the schedule of traffic signals ahead. Results from two deployments of SignalGuru, using iPhones in cars in Cambridge (MA, USA) and Singapore, show that traffic signal schedule can be predicted with very good accuracy. On average, SignalGuru comes within 0.66<italic>s</italic>, for pre-timed traffic signals and within 2.45<italic>s</italic>, for traffic-adaptive traffic signals. Feeding SignalGuru's predicted traffic schedule to our Green Light Optimal Speed Advisory (GLOSA) application, our vehicle fuel consumption measurements show savings of 20.3&#37;, on average. SignalGuru information can also be fed into several other envisioned applications to further improve fuel efficiency, vehicle flow, travel time and road safety. The example of SignalGuru illustrates that with collaboration and adaptation, mobile device-based computing platforms can support a rich set of challenging services without the help of cloud servers and the associated long-range communications. Overall, this thesis advocates and demonstrates that, with collaboration and adaptation, mobile devices can effectively support a rich set of services and thus reduce the need for slow and costly long-range cellular communications to cloud servers. Several traditional cloud services that operate on very large amounts of data can be selectively and adaptively hosted on mobile devices. Furthermore, novel mobile services that may seem prohibitively resource-intensive and challenging can be enabled and hosted on confederations of collaborating mobile devices. In this way, the mobile user experience can be greatly improved and a significant amount of the increasingly scarce long-range communication bandwidth can be saved.en_US
dc.language.isoenen_US
dc.publisherPrinceton, NJ : Princeton Universityen_US
dc.relation.isformatofThe Mudd Manuscript Library retains one bound copy of each dissertation. Search for these copies in the <a href=http://catalog.princeton.edu> library's main catalog </a>en_US
dc.subjectadaptationen_US
dc.subjectarchitecturesen_US
dc.subjectcollaborationen_US
dc.subjectmiddlewareen_US
dc.subjectmobile devicesen_US
dc.subjectservicesen_US
dc.subject.classificationComputer engineeringen_US
dc.subject.classificationComputer scienceen_US
dc.titleCollaborative and Adaptive Mobile Device-resident Service Architecturesen_US
dc.typeAcademic dissertations (Ph.D.)en_US
pu.projectgrantnumber690-2143en_US
Appears in Collections:Electrical Engineering

Files in This Item:
File Description SizeFormat 
Koukoumidis_princeton_0181D_10108.pdf5.72 MBAdobe PDFView/Download


Items in Dataspace are protected by copyright, with all rights reserved, unless otherwise indicated.