1MIST: Cellular Data Network Measurementfor Mobile ApplicationsMike P. Wittie, Brett Stone-Gross, Kevin C. Almeroth and Elizabeth M. BeldingDepartment of Computer Science, University of California, Santa BarbaraSanta Barbara, CA 93106-5110{mwittie, bstone, almeroth, ebelding}@cs.ucsb.eduAbstract— The rapid growth in the popularity of cellularnetworks has led to aggressive deployment and a rapid expansionof services. Services based on the integration of these networksinto the Internet have only recently become available, but areexpected to become very popular. One current limitation to thedeployment of many of these services is poor or unknown networkperformance, particularly in the cellular portion of the network.Our goal in this paper is to motivate and present the Mobile Inter-net Services Test (MIST) platform, a new architecture to measureand characterize cellular network performance as experiencedby mobile devices. We have used MIST to conduct preliminarymeasurements; evaluate its effectiveness; and motivate furthermeasurement research.I. INTRODUCTIONRecent years have seen a rapid expansion of data services incellular networks. As aggressive competition between cellularservice providers has lead to decreases in average per-userrevenue, network providers began to look at ways to increasenon-voice revenue [1]. Studies of customer network usagehave indicated the propensity of customers, especially in theUS, to adopt new network services hinting at the possibility ofincreasing data usage via heavier application traffic [2]. Thefuture is also expected to bring an ever increasing integrationof various multimedia, packet, and location-based cellular datanetwork services. These services will certainly increase thedemands on these networks [2], [1].The increased customer and service provider focus ondata services has attracted the interest of mobile applicationdevelopers ready to cash in on a new trend. These companieshave recognized the opportunity to create revenue generatingvalue chains between themselves and the network providers,and therefore, are rapidly exploring the market space [3].Popular mobile applications, well supported by the networkinfrastructure, benefit both the network providers and applica-tion developers by driving both data service usage, as well asrevenue from application downloads and service fees. For anapplication to be successful, however, it needs to be broughtto the market quickly and offer a high quality service. Beatingthe competition even by a week can make a major differencein the level of application adoption by the user community.The high quality of an application, equally important to itssuccess as time to market, needs to be assured for a largenumber of devices and network provider combinations [4].Studies of the mobile application development processeshave identified a large number of challenges to applicationdevelopment time and interoperability [1], [5], [6], [4]. Twomain challenge areas have been identified. First, while JAVA isbecoming the preferred language of mobile application devel-opers, various mobile JAVA Virtual Machine implementationsare riddled with inconsistencies, necessitating custom codeoptimizations for many platforms [5]. Second, the lack ofinformation about the behavior and performance of differentnetwork protocols on different devices and network providerscomplicates mobile application development [4]. In addition,these performance issues also affect the ability to optimizemany application types [6]. While the incompatibilities ofJAVA are a problem known throughout the industry, and onethat is being addressed, the detailed characterization of thecellular environment as seen by mobile devices remains largelyunclear.In spite of the growing interest in cellular data networkperformance, from the point of view of mobile device perfor-mance, the area remains largely unexplored. Most cellular datanetwork studies have investigated the performance of particu-lar network technologies [7], [8], [9], [10], [11], focusing onnetwork infrastructure optimization and provisioning. Besidesthe underlying network technology, there are many factorsaffecting network performance, like radio technology, phonehardware, or network settings. Additionally, it is difficult toextrapolate the current state of technology due to staleness ofresults or difficulties in comparison across different studies. Toaid mobile application developers and identify performancepitfalls on cellular data networks, a number of studies havefocused on the relationship between these networks and thetransport protocols used by the applications [12], [13]. Whilethese transport protocol studies did offer insight to the net-working community, other studies have shown that transportprotocols optimized for the wireless environment, more oftenthan not, do not translate to application performance improve-ments [14]. Indeed, the best way to assure good applicationusability and performance is to customize application code tothe specific platform and network characteristics.In this paper we propose the Mobile Internet ServicesTest (MIST) platform, an architecture designed to characterizecellular data network performance as experienced by individ-ual mobile devices. The measurement network performance,network technology, and mobile hardware combinations ex-poses the effects of the interactions and particularities ofthe underlying technologies. Since mobile applications areexpected to work on a wide variety of platforms, understanding2the performance variations is crucial to making applicationdesign decisions, thus assuring high application adoption rateswithin the user community.MIST is comprised of a mobile application connected toa server back-end. A number of network performance testsare performed between the mobile application and the MISTservers in order to assess network latency, jitter, throughput,and various timeout intervals. Each set of measurement datais saved in a database along with the network and mobiledevice configuration information used in the test. MIST islightweight, stateless, and highly scalable. Most importantly,its deployment does not require changing or augmenting thecellular data network infrastructure. It can be deployed on topof mobile devices themselves, configured to measure exactlythe network characteristics a mobile application developer isinterested in, and report directly on the user experience.Using a simple set of test data collected by MIST,