Performance Evaluation of a Live, Crowdsensing Based Transit Feed Service Architecture K´ aroly Farkas * , R´ obert Szab´ o † and Bern´ at Wiandt * * Dept. of Networked Systems and Services, † HSNLab, Dept. of Telecommunications and Media Informatics, Budapest University of Technology and Economics, Budapest, Hungary Email: farkask@hit.bme.hu, robert.szabo@tmit.bme.hu, bwiandt@hit.bme.hu Abstract—Taking public transportation is an efficient and environmentally sound way of traveling in most of the cities. Unfortunately, the static schedule information of the public transport vehicles, available at the stops, on the web or in some special format like GTFS (General Transit Feed Specification), usually does not reflect the actual traffic situation. However, real- time traffic updates require the gathering of an immense amount of tracking data. Mobile crowdsensing, via the mobile devices of the crowd, can offer a cheap and efficient way for collecting such data. Nonetheless, for motivating active participation some day zero service must be provided to the users. In this paper, we discuss how to realize a live public transport information service extending a static timetable based on GTFS data, as the day zero service, using the power of the crowd for data collection. We detail the design of such a service implemented by an XMPP (Extensible Messaging and Presence Protocol) based mobile crowdsensing architecture and evaluate its performance. We show how we can build a scalable service architecture even with commodity hardware to handle thousands of users. Keywords—Crowdsensing, Public transport, GTFS, Pub- lish/subscribe, XMPP I. I NTRODUCTION Commuting has become part of our daily routine which results in spending a significant part of our time with traveling. Public transportation offers a viable alternative for commuters, especially in crowded cities, reducing private vehicle usage, fuel consumption, environmental pollution and alleviating traf- fic congestion. However, it is important for the passengers to have accurate information about the arrival time at the stops of the public transit vehicles. Otherwise, unexpectedly long waiting times can frustrate passengers and divert them from using public transport. It is even better if some other extra information about the arriving vehicle, for instance the congestion level or baby buggy friendliness, is also provided, making travel planning easier. Fortunately, most transit operators of big cities make their timetables or transit feeds freely available for the passengers or third party application developers. The General Transit Feed Specification (GTFS) [1] – originally developed by Google – evolved into the de-facto standard to exchange transit feeds publicly. However, the widely used GTFS format enables only the exchange of static transit information (e.g., departure schedules, travel time, operating hours), which does not reflect live traffic conditions. Remedy to live updates is the relatively new real-time extension to GTFS [2], which relies on operators to invest into a real-time fleet tracking and communication infrastructure to be able to provide live service updates. Unfortunately, nowadays very few operators offer such services due to the necessary investments. Another approach to live updates is using mobile crowd- sensing 1 [3] and let the crowd collect the information re- quired to extend the basic timetable service. In this case, voluntary passengers sense and send live service updates (e.g., delays, congestion, hazard, etc.) to a service provider via their smart-phones. The service provider then aggregates, cleans, analyzes the collected data, and derives and disseminates the live updates to the users. For sensing, the built-in and ubiquitous sensors of the mobile phones can be used either in a participatory or an opportunistic way depending on whether data collection happens with or without user involvement. The contribution of every traveler can be useful. Hence, passengers waiting for a trip can report the line number with a timestamp of every arriving public transport vehicle at a stop during the waiting period. On the other hand, on-board passengers can send actual location information of the moving vehicle periodically and report events of arrival at/departure from the stops. Although the participatory sensing based approach is a viable alternative, it faces several challenges. The basic chal- lenge, similarly to other crowdsensing based services, is how passengers can be motivated to participate in data collec- tion. We believe, that a day zero service, which is provided from the beginning and improved, based on the crowd-sensed data, following incremental service development, can be an appropriate incentive. Along this line, as a sequel work to [4] and [5], we propose and investigate a scenario, where a static transit feed as the day zero service is improved and updated with live service updates from participatory users. We show how such a scenario can be implemented by the Extensible Messaging and Presence Protocol (XMPP) [6]. We evaluate different XMPP server setups under load. We present measurement results showing that even commodity PCs are able to carry on with the load and handle several thousand users, and that XMPP server clustering can remedy service scaling. We show, that with our architecture and a GTFS emulator static and live service updates can easily be combined introducing incremental service improvements based on mobile participatory users. The rest of the paper is structured as follows. In Sec. II we 1 We use the terms crowdsensing, crowdsourcing and participatory sensing interchangeably in this paper. – 251 – INES 2014 • IEEE 18th International Conference on Intelligent Engineering Systems • July 3-5, 2014, Tihany, Hungary 978-1-4799-4615-0/14/$31.00 ©2014 IEEE