GAIUS: A New Mobile Content Creation And DiffusionEcosystem For Emerging Regions

Talal AhmadNew York Universityahmad@cs.nyu.edu

Yasir ZakiNYU Abu Dhabi

yasir.zaki@nyu.edu

Thomas PötshNYU Abu Dhabi

thomas.poetsch@nyu.edu

Jay ChenNYU Abu Dhabijchen@cs.nyu.edu

Arjuna SathiaseelanGAIUS Networks

arjuna.sathiaseelan@gaiusnetworks.com

LakshminarayananSubramanian

New York Universitylakshmi@cs.nyu.edu

ABSTRACTDespite increasing mobile Internet penetration in emerging regions,the growth of web page complexity combined with inadequatecontent delivery mechanisms and lack of relevant local contentmake the web experience poor for users in these regions. In thispaper, we propose GAIUS, a content ecosystem that enables theefficient creation and dissemination of locally relevant web content.GAIUS aims at creating a community content exchange that enablesusers to easily create and consume relevant web content with thehelp of a mobile application. Based on an early deployment ofGAIUS in various regions, we show that our ecosystem enableslocalized content creation and diffusion efficiently.

CCS CONCEPTS• Networks → Location based services; World Wide Web (net-work structure); Public Internet;

KEYWORDSMobile, Content, LocalACM Reference Format:Talal Ahmad, Yasir Zaki, Thomas Pötsh, Jay Chen, Arjuna Sathiaseelan,and Lakshminarayanan Subramanian. 2019. GAIUS: A New Mobile ContentCreation And Diffusion Ecosystem For Emerging Regions. In Tenth Inter-national Conference on Information and Communication Technologies andDevelopment (ICTD ’19), January 4–7, 2019, Ahmedabad, India. ACM, NewYork, NY, USA, 5 pages. https://doi.org/10.1145/3287098.3287130

1 INTRODUCTIONMany challenges constrainweb access in developing regions. Amea-surement study by Zaki et. al. [20] showed that developing regionslike Ghana suffer from high page load times (multi-seconds), dueto HTTP redirects, DNS lookups and TLS/SSL connection setups.They found the actual time spent downloading content represents

Permission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for components of this work owned by others than ACMmust be honored. Abstracting with credit is permitted. To copy otherwise, or republish,to post on servers or to redistribute to lists, requires prior specific permission and/or afee. Request permissions from permissions@acm.org.ICTD ’19, January 4–7, 2019, Ahmedabad, India© 2019 Association for Computing Machinery.ACM ISBN 978-1-4503-6122-4/19/01.https://doi.org/10.1145/3287098.3287130

only a small fraction of the end-to-end page download time. Kora-dia et. al. [17] have shown that cellular data connectivity in Indiasuffers from significantly high latencies. Other works have pointedout traffic engineering and lack of infrastructure as reasons forhigh delays [5, 11, 14, 15]. In extreme cases, power outages canalso lead to network disruptions [3]. To overcome the challengesof poor web performance in developing countries, several systemsand techniques have been proposed in recent years to optimizeweb browsing over poor network connections including network-level optimizations, caching techniques and content distributionmechanisms [6–9, 16, 19]. In addition to this, Content DistributionNetworks (CDNs), which traditionally aim to replicate relevantcontent in the network edges [1, 2], are ineffective for enhancingperformance in the developing world [4, 12, 18]. The underlyingfactors are three-fold: poor connectivity [10], high latencies, andweb complexity.

Beyond the well-known technical and infrastructural issues, onemajor problem that has not been explored in detail is the lack ofcheap, locally relevant content. The vast majority of web pageson the Internet are from developed countries, far removed fromaudiences in developing regions. This shortage of relevant localcontent makes the Internet less attractive for users in developingregions. The existing web ecosystem alsomakes it hard for people topublish locally relevant content. Furthermore, bloated web contentand remote hosting also undermine web performance and userexperience.

To address these problems, we propose GAIUS, a mobile ecosys-tem for local content creation and diffusion. Beyond localized con-tent creation and diffusion, GAIUS also improves localized webperformance and provides support for local advertisers. The GAIUSecosystem is an immense improvement over the status quo in thecontext of localized content.

Architecturally, GAIUS consists of a set of locally distributedservers that host GAIUS pages (interchangeably referred to as chan-nels) and an application running on a user’s smartphone devicethat enables consumption and creation of local content. The GAIUSpages are represented in a highly concise language format insteadof HTML and JavaScript, to minimize communication overhead.Finally, the GAIUS app on an end user’s device receives, interprets,and renders the pages. The GAIUS application is also used to createcontent and publish it on the GAIUS servers. Together, these com-ponents enable the easy creation, diffusion, and consumption of

https://doi.org/10.1145/3287098.3287130https://doi.org/10.1145/3287098.3287130

ICTD ’19, January 4–7, 2019, Ahmedabad, India T. Ahmad, Y. Zaki, T. Pötsh, et al.

content at low cost. In summary, the GAIUS ecosystem is designedaround a number of key concepts:• Hosting and diffusing content close to the users.• Using a novel specification language.• Coordinating a mobile application and an edge server toefficiently view and create content.

We have preliminary deployment of GAIUS in a few small com-munities around theworld. Our early deployment shows that GAIUSenables users to create and publish pages without extensive trainingor web publishing experience.

2 GAIUS ECOSYSTEMThe GAIUS ecosystem solves the problem of content delivery fromlocal and global content/advertisement providers to mobile users.The main features of GAIUS are:• GAIUS enables local content creation and diffusion by allow-ing users to create content within a mobile application.• GAIUS content is defined using our own specification lan-guage called MAML, instead of JavaScript and HTML. Thismakes the pages simpler and makes them load faster• GAIUS provides a new way to identify and address con-tent. In GAIUS content is divided into channels, where eachchannel is maintained by a single users or content provider.Within the channel a user/content provider can create sev-eral pages of his/her liking.• GAIUS introduces a new ad ecosystem by providing supportfor local ads. Different tiers of advertisers can be accom-modated in such an ecosystem varying from small localindividuals to large corporations.

The GAIUS ecosystem as shown in Figure 1 comprises a set ofdistributed edge servers that help generate and serve different localpages at the edge based on user requests. Each edge is envisionedto serve a ‘region’ (e.g., a city). One proposal for deployment is toplace a edge servers at a packet gateway level in a cellular corenetwork.

GAIUS WAN Platform

Global Ad Provider

GAIUS Edge Cloud Platform

GAIUS Edge Cloud Platform

Cloudlet Google

Cloudlet Facebook

Cloudlet Youtube

Cloudlet Instragram

Cloudlet school

Cloudlet community radio

Cloudlet community TV

Cloudlet Municipality

Cloudlet Google

Cloudlet Facebook

Cloudlet Youtube

Cloudlet Instragram

Cloudlet school

Cloudlet community radio

Cloudlet community TV

Cloudlet Municipality

Local content Providers

Local Ad Providers

Local content Providers

Local Ad Providers

Figure 1: A high level view of the GAIUS ecosystem

2.1 Mobile App Markup Language (MAML)In GAIUS, we use a novel way to define content, we call it MobileApp Markup Language and refer to it as MAML in the rest of thepaper. In regular HTML pages, the complexity of rendering a web

page after the downloading of the first index.html is dictated bythe different DNS resolutions, redirects and inherent complexity inthe object dependency tree embedded in the web page. In contrast,MAML removes this complexity by simplifying the way a web pageis written. This makes the rendering of the MAML pages extremelysimple and enables prioritization of object downloads.

To retain functional compatibility, MAML pages must be sim-ilar in look and feel as conventional HTML web pages. Conse-quently, the current MAML specification supports six differentobject types: image, text, video, rectangle, text-field, and button. Forexample, the MAML image object represents the mapped HTMLimage tag, which contains the URL of the actual image that needs tobe prefetched from the network. The image object has the currentattributes: URL, x and y coordinates, width and height of the image,and an optional hyperlink in case the image is click-able.

MAML object specification incorporates styling and interactivity,thus eliminating the need for JavaScript and CSS. This helps cutdown the size of the page and reduces page load time. Upfrontspecification of web objects in a MAML page enables prioritizationof downloads and eliminates recursive requests while renderinga page. The flattened tree structure simplifies the object loading,object parsing, and the object rendering, thus enabling faster pagerendering.

2.2 GAIUS ChannelsThe legacy Internet relies on the notion of Uniform Resource Loca-tors (URLs), where users type in the URL in a web browser to accessthe associated page. This requires the browser to do a DNS lookupto map the requested URL into the IP address of the machine thathosts the information. GAIUS defines a new way for users to accessand search for content, that is the notion of information channels.A channel is associated with a content provider and contains spe-cific information that they want to disseminate to users. A channelcan contain multiple sub-pages each with a unique topic or setsof topics. These topics are used to help users search for relevantinformation. Users can also subscribe to channels to easily viewnewly generated content.

The local ecosystem takes input from different stake holders(i.e. content providers, global and local advertisers, and users) andimplements on-the-fly page creation. Figure 1 shows the inputs ofthe GAIUS ecosystem. From these inputs, GAIUS generates a webpage and any associated advertisements for the user. The user’sinterface consists of a mobile application which enables user toconnect to an edge and request web pages and information channels.The GAIUS app displays the web page using MAML instead ofJavaScript and HTML.

2.3 GAIUS Mobile AppMobile phones and mobile-broadband are among the top used ac-cess technologies for users in developing regions to get on theInternet. While GAIUS leaves the door open for content providersto optimize their pages, GAIUS also gives users more control overtheir web experience and allows them to: (i) view and search forinformation channels and content, (ii) control the fidelity of the

GAIUS: A New Mobile Content Creation & Diffusion Ecosystem ICTD ’19, January 4–7, 2019, Ahmedabad, India

requested content, thus controlling their data consumption, (iii) cre-ate their own local content and distribute it within the community,and (iv) advertise within their community at low cost.

Content Creation: The GAIUS ecosystem changes how con-tent is created by allowing users to generate their own contentusing predefined templates that do not require significant technicalknowledge. Templates can be filled with text, images, and videosthrough a simple WYSIWYG interface, where users can create theirown information channel hosted at the nearest GAIUS server. Oncethe channel is published, all users within the community are ableto browse through this generated content. Since all of the localcontent is hosted nearby, the hosting and bandwidth costs can beminimized. GAIUS relies on the local ad marketplace to cover thecosts of local content.

2.4 Content Policy SpecificationsOne of the motivations behind the GAIUS ecosystem is to serveusers different versions of the same content dictated by policy in-puts. For example, based on the user budget, he/she can requesta version of a page that is optimized for lower data consumption.However, user-side economics is not the sole determining factor, anumber of additional parameters also play a role in page creation.The page layout and the content dependencies are typically de-termined by the content provider and the priority of the content.Hence, content providers specify the overall layout of the page aswell as how the content is related and structured within the page.On the other hand, advertisers determine which ads needs to bedisplayed to the user based on their own economic considerations(e.g., relevance to the users, revenue, advertisers budget). To fastpage loads, the users’ network conditions also play a role. All ofthese factors are combined to produce the final page.

2.5 AdvertisementsGAIUS is a collection of highly localized ecosystems where eachecosystem has 100-100k users, so our ad model is completely differ-ent from conventional content providers. The GAIUS applicationallows individual local users to easily publish their own advertise-ments in form of videos, images or text from their phone. Thead is then added to the local repository to be displayed to userswithin their community. In GAIUS we implement the aggregatorslogic(how relevant and ad is for a user) as part of our edge logic. Atthe edge, ads are chosen based on location, content providers input,and users interests. Unlike conventional aggregators that collect afixed fraction of ad revenue generated [13], we plan to introducepricing that won’t burden the local advertisers.

3 IMPLEMENTATION3.1 GAIUS Edge ServerThe GAIUS edge server is based on the Apache Server implemen-tation, using the the Common Gateway Interface (CGI). When aspecific user request a MAML index file, Apache redirects the re-quest to our GAIUS policy engine implementation. Based on variousspecifications, the policy engine creates and delivers a custom ver-sion of the requested page. In addition, the GAIUS ad selectionpolicy is responsible to mix the local content with relevant localadvertisements from the ad market place.

The edge server maintains a list of the information channelsit hosts. This is used to answer user queries about the availableinformation channels. Based on a user’s request, a specific versionthat matches the user preferences is served to the user. In our pro-totype we emulate content providers by creating multiple versionsof several popular websites, where each version contains differentimage sizes by manipulating their resolution. Figure 2 shows twoversions of the CNN web page with different fidelity levels (highvs low resolution). The figure also shows the overall page size thatis significantly smaller for the lower resolution version, i.e. about80% reduction in size.

(a) High resolution (377 kB) (b) Low resolution (73 kB)

Figure 2: GAIUS rendered CNNweb page with different userfidelities

3.2 GAIUS Mobile AppFor end-users we built the GAIUS Android app shown in Figure 3.The app has a simple user interface with multiple features: chan-nel browsing, content creation/editing, ad creation. The channelbrowsing feature displays the different information channels tothe user sorted based on a mixture of the content proximity to theuser and the content popularity. The list displays the channel titleand icon for each information channel. When the user clicks ona specific channel, the app requests the corresponding informa-tion channel. This can lead to one of the following: if the channelconsists of a single page then the corresponding MAML page issent to the user and is rendered by the app browser; on the otherhand, if the information channel consists of several sub-pages thena list of sub-pages is sent to the app from the edge server and isdisplayed to the user to choose from. The GAIUS app user is able toset his own fidelity level preference between three different levels:high, medium and low. Upon setting the fidelity level, the corre-sponding image quality from the edge server changes. In addition,the edge server changes the type of the ads within the requested

ICTD ’19, January 4–7, 2019, Ahmedabad, India T. Ahmad, Y. Zaki, T. Pötsh, et al.

MAML pages, from video ads at the high fidelity level, images atthe medium fidelity, and plain text ads at the low fidelity level.

Figure 3: GAIUS Android application interface

4 EARLY DEPLOYMENTIn this section, we discuss our early deployment experiences. Wedeployed the GAIUS ecosystem in several communities in differentcountries. We specifically targeted small scale deployments in selectcommunities. The deployments consisted of a single GAIUS server,where the content was hosted, and a number of smart-phone usersfrom different communities who volunteered to test out our system.The users were asked to download the GAIUS Android applicationand start using it to browse various channels, and to create theirown local content and advertisements. We currently have about 38active users that span across 10 different countries. The server hostsmore than 55 information channels, some of them were created byus, while others were created by our users.

To inspire the newly joined users and in order to have localrelevant content for them to browse through, we pre-populatedour server with local content specific to user locations. This initialcontent was created by translating popular websites and RSS feedswithin the users’ region. The content was chosen from a widevariety of categories ranging from news, sports, entertainment,events, food, etc.

To evaluate the current deployment of the GAIUS ecosystem,we record statistics for each individual page request in GAIUS. Thefollowing statistics were recorded: the overall page load time, theoverall page size, the page fidelity requested by the user, the user’sgeographic-location, the user’s network type and mobile phonemodel.

Figure 4a shows the page load time box plot for the overall pagerequests as well as for the different individual fidelity levels. Thex-axis shows the fidelity level, whereas the y-axis displays the pageload time. It can be seen that the median page load time is about1.6 seconds, with the 75% not exceeding 3 seconds. In addition, itcan be seen that the "high" fidelity level has the highest median

page load time, in contrast to the "medium" fidelity which has alower median value, and the "low" fidelity having a median pageload time below one second.

(a) GAIUS page load time (s) (b) GAIUS page size (kB)

Figure 4: Box plot showing the distribution of page loadtimes and page size for different fidelity levels.

Figure 4b shows the page size box plot again overall and acrossthe multiple fidelity levels. The figure clearly shows how effectiveGAIUS is in reducing the overall page size with a median smallerthan 40kB. The figure also shows how the medium and low fidelitymanages to reduce the overall page load time by more than 6x.

Figure 5 shows the distribution of how page load times variedas a function of network used to access the pages. The lowest pageload times were experienced by users connected over WiFi whichhas the lowest median value. Looking at median values, we can seethat users connected via 4G and 3G networks experienced page loadtimes slightly more than WiFi but still in the acceptable range offew seconds. The 2G network had the worst performance and had apage load time of around 10 seconds but this is still an improvementover status quo.

In summary, this paper shows that the GAIUS ecosystem canimprove web experience in emerging regions by providing efficientmechanism for localized content creation and diffusion. GAIUSenables this by leveraging a mobile application and a novel con-tent specification language. We observed that GAIUS content loadsfaster than conventional web content under various network con-ditions.

Figure 5: Box plot showing the distribution of page loadtimes for different network types.

GAIUS: A New Mobile Content Creation & Diffusion Ecosystem ICTD ’19, January 4–7, 2019, Ahmedabad, India

ACKNOWLEDGMENTSWe thank the Cisco Research Grant for partially supporting Laksh-minarayanan Subramanian, and Talal Ahmad on this project. Wewould also like to thank the anonymous referees for their valuablecomments and helpful suggestions.

REFERENCES[1] 2017. Akamai Content Delivery Network. (2017). http://www.akamai.com/[2] 2017. Cloudflare Content Delivery Network. (2017). http://www.cloudflare.com/[3] Talal Ahmad, Shankar Kalyanaraman, Fareeha Amjad, and Lakshmi Subramanian.

2015. Solar vs Diesel: Where to draw the line for cell towers?. In Proceedings ofthe 7th International Conference on Information and Communication Technologiesand Development, ICTD 2015, Vol. 15. Association for Computing Machinery. DOI:https://doi.org/10.1145/2737856.2738028

[4] Zachary S. Bischof, John P. Rula, and Fabián E. Bustamante. 2015. In and Outof Cuba: Characterizing Cuba’s Connectivity. In Proceedings of the 2015 InternetMeasurement Conference (IMC ’15). ACM, New York, NY, USA, 487–493. DOI:https://doi.org/10.1145/2815675.2815718

[5] Josiah Chavula, Nick Feamster, Antoine Bagula, and Hussein Suleman. 2015.Quantifying the Effects of Circuitous Routes on the Latency of Intra-Africa InternetTraffic: A Study of Research and Education Networks. 64–73. DOI:https://doi.org/10.1007/978-3-319-16886-9_7

[6] Jay Chen, David Hutchful, William Thies, and Lakshminarayanan Subramanian.2011. Analyzing and Accelerating Web Access in a School in Peri-urban India. InProceedings of the 20th International Conference Companion on World Wide Web(WWW ’11). ACM, New York, NY, USA, 443–452. DOI:https://doi.org/10.1145/1963192.1963358

[7] Jay Chen, Russell Power, Lakshminarayanan Subramanian, and Jonathan Ledlie.2011. Design and Implementation of Contextual Information Portals. In Proceed-ings of the 20th International Conference Companion on World Wide Web (WWW’11). ACM, New York, NY, USA, 453–462. DOI:https://doi.org/10.1145/1963192.1963359

[8] Jay Chen, Lakshminarayanan Subramanian, and Jinyang Li. 2009. RuralCafe: WebSearch in the Rural Developing World. In Proceedings of the 18th InternationalConference on World Wide Web (WWW ’09). ACM, New York, NY, USA, 411–420.DOI:https://doi.org/10.1145/1526709.1526765

[9] Marshini Chetty, David Haslem, Andrew Baird, Ugochi Ofoha, Bethany Sumner,and Rebecca Grinter. 2011. Why is My Internet Slow?: Making Network SpeedsVisible. In Proceedings of the SIGCHI Conference on Human Factors in ComputingSystems (CHI ’11). ACM, New York, NY, USA, 1889–1898. DOI:https://doi.org/10.1145/1978942.1979217

[10] Marshini Chetty, Srikanth Sundaresan, Sachit Muckaden, Nick Feamster, andEnrico Calandro. 2013. Measuring Broadband Performance in South Africa. InProceedings of the 4th Annual Symposium on Computing for Development (ACMDEV-4 ’13). ACM, New York, NY, USA, Article 1, 10 pages. DOI:https://doi.org/10.1145/2537052.2537053

[11] Rodérick Fanou, Pierre Francois, and Emile Aben. 2015. On the Diversity of Interdo-main Routing in Africa. 41–54. DOI:https://doi.org/10.1007/978-3-319-15509-8_4

[12] Rodérick Fanou, Gareth Tyson, Pierre Francois, and Arjuna Sathiaseelan. 2016.Pushing the Frontier: Exploring the African Web Ecosystem. In World Wide WebConference (WWW).

[13] Phillipa Gill, Vijay Erramilli, Augustin Chaintreau, Balachander Krishnamurthy,Konstantina Papagiannaki, and Pablo Rodriguez. 2013. Best Paper – Followthe Money: Understanding Economics of Online Aggregation and Advertising.In Proceedings of the 2013 Conference on Internet Measurement Conference (IMC’13). ACM, New York, NY, USA, 141–148. DOI:https://doi.org/10.1145/2504730.2504768

[14] J Gilmore, N Huysamen, and A Krzesinski. 2007. Mapping the african internet.In Proceedings Southern African Telecommunication Networks and ApplicationsConference (SATNAC), Mauritius.

[15] Arpit Gupta, Matt Calder, Nick Feamster, Marshini Chetty, Enrico Calandro, andEthan Katz-Bassett. 2014. Peering at the internet’s frontier: A first look at ispinterconnectivity in Africa. Passive Active Measurement Conference (PAM) (2014),204–213.

[16] Sibren Isaacman and Margaret Martonosi. 2009. The C-LINK System for Collabo-rative Web Usage: A Real-World Deployment in Rural Nicaragua. (2009).

[17] Zahir Koradia, Goutham Mannava, Aravindh Raman, Gaurav Aggarwal, VinayRibeiro, Aaditeshwar Seth, Sebastian Ardon, AnirbanMahanti, and Sipat Triukose.2013. First Impressions on the State of Cellular Data Connectivity in India. InProceedings of the 4th Annual Symposium on Computing for Development (ACMDEV-4 ’13). ACM, New York, NY, USA, Article 3, 10 pages. DOI:https://doi.org/10.1145/2537052.2537064

[18] Asheesh Sharma, Manveen Kaur, Zahir Koradia, Rahul Nishant, Sameer Pandit,Aravindh Raman, and Aaditeshwar Seth. 2015. Revisiting the State of CellularData Connectivity in India. In Proceedings of the 2015 Annual Symposium onComputing for Development (DEV ’15). ACM, New York, NY, USA, 149–157. DOI:https://doi.org/10.1145/2830629.2830649

[19] William Thies, Janelle Prevost, Tazeen Mahtab, Genevieve T. Cuevas, SaadShakhshir, Alexandro Artola, Ro Artola, Binh D. Vo, Yuliya Litvak, SheldonChan, Sid Henderson, Mark Halsey, Libby Levison, and Saman Amarasinghe.2002. Searching the World Wide Web in Low-Connectivity Communities. (2002).

[20] Yasir Zaki, Jay Chen, Thomas Pötsch, Talal Ahmad, and LakshminarayananSubramanian. 2014. Dissecting Web Latency in Ghana. In Proc. of the ACMInternet Measurement Conference (IMC). Vancouver, BC, Canada.

http://www.akamai.com/http://www.cloudflare.com/https://doi.org/10.1145/2737856.2738028https://doi.org/10.1145/2815675.2815718https://doi.org/10.1007/978-3-319-16886-9_7https://doi.org/10.1007/978-3-319-16886-9_7https://doi.org/10.1145/1963192.1963358https://doi.org/10.1145/1963192.1963358https://doi.org/10.1145/1963192.1963359https://doi.org/10.1145/1963192.1963359https://doi.org/10.1145/1526709.1526765https://doi.org/10.1145/1978942.1979217https://doi.org/10.1145/1978942.1979217https://doi.org/10.1145/2537052.2537053https://doi.org/10.1145/2537052.2537053https://doi.org/10.1007/978-3-319-15509-8_4https://doi.org/10.1145/2504730.2504768https://doi.org/10.1145/2504730.2504768https://doi.org/10.1145/2537052.2537064https://doi.org/10.1145/2537052.2537064https://doi.org/10.1145/2830629.2830649

Abstract1 Introduction2 GAIUS Ecosystem2.1 Mobile App Markup Language (MAML)2.2 GAIUS Channels2.3 GAIUS Mobile App2.4 Content Policy Specifications2.5 Advertisements

3 Implementation3.1 GAIUS Edge Server3.2 GAIUS Mobile App

4 Early DeploymentAcknowledgmentsReferences