2.
RELATED WORK
Hidden costs associated with current mobile ads deliv- ery process were highlighted by a number of groups. Khan et al. [16] and Zhang et al. [29] reported that, 30-60min daily use of popular apps such as fruit ninja and angry birds can consume up to 40-50MB of data per month. Vallina- Rodriguez et al. [27] measured that for 3% of Android users and iPhone users daily ad traffic is more than 1MB and 3MB respectively, using a large data set obtained from a major European mobile operator.
Pathak et al. [19] showed that 65%-70% of the energy consumption of free apps is due to advertisements. Fur- thermore, it has been shown that ad refresh rates keeps the smartphone always in the high power consuming RRC states even though the core app functionality does not require any network communication [21].
Number of previous work proposed several solutions for mobile ads delivery process. However most of those looked at only one or two out of three requirements. Due to the conflicting nature of the requirements, such approaches do not fulfil the remaining requirement or in some cases push it to an unacceptable level.
In order to minimize resource consumption Khan et al. proposed CAMEO [16] in which users’ context is predicted and relevant ads are pre-fetched to the mobile device using low cost and energy efficient WiFi networks. While such ap- proach saves the bandwidth and smartphone energy it raises privacy concerns. All the user information will be passed to the ad-network enabling user profiling. Adcache [27] is a sim- ilar caching solution and the privacy aspects were addressed by decoupling the permission required for ad SDKs by the permissions required for the core app operation. Restricting the information given to the ad-network reduces the overall usefulness as ad networks cannot provide targeted ads with-out information and as a result possibility of user clicking an ad reduces. AdSplit [23], Addroid [20] also proposed similar permission separation.
Haddadi et al. proposed MobiAd [13] to provide location based advertising whilst preserving privacy. The user de- vice will select ads from a pool of broadcasted ads based on user interests. As pointed out in [22] such approach is suitable only for smaller ad inventories due to limitations in broadcast channels, hence low usefulness.
Solutions proposed for online advertising domain, Pri- vad [12] and Adnostic [26] tried to address the problem by downloading advertisements using broader categories (e.g Sports) and deciding the fine-grained targeting (e.g. Sports- >Cricket) at the client side. Though this addresses the pri- vacy requirement, it’s not directly suitable for smartphone environment due to high resource consumption.
MASTADs [22] proposed to use opportunistic communi- cation to anonymously propagate user interests to a semi- trusted broker and to pre-fetch the advertisements through the same means. This approach might consume smartphone energy as smartphones directly involve in forwarding each others’ interests and advert creatives collaboratively.
Morerpiv [6] presented the idea of the use of OS supported personalization service using “personas”. However the archi- tecture did not elaborate how the “profile unlinkability” can be achieved. Further the statistics and reporting require- ments of the mobile advertising were also not addressed.
Recent work by Mohan et al. [18] studied the energy vs revenue impact of ads pre-fetching and showed that energy overhead can be reduced by 50% using pre-fetching intervals of 20 minutes. However they did not address the privacy requirement in their solution.
Several work [3, 24] looked at the possibility of offload- ing smartphone processing to nodes with unconstrained re- sources. In [4] Aceres et al. presented the concept of individ- ually owned personal servers as a means to increase privacy in online social networks and location privacy. These works did not address the unique requirements of mobile advertis- ing. Several older work such as [2, 15] looked at the use of middleboxes for tailoring services, to the device and network characteristics.
We differentiate our work form the previous work by pro- viding an end-to-end solution to the mobile ads distribution while addressing the requirements of minimum resource con- sumption, minimizing the loss of privacy and usefulness.