190 likes | 334 Vues
Follow the Crowd: On QoE for Internet Applications . Tobias Hoßfeld. www3.informatik.uni-wuerzburg.de www.t-hossfeld.de. What is the Internet crowd consuming ?. Web and Cloud Applications Online Video, Web Browsing, Downloads, Cloud Services, etc. Why relevant?
E N D
Follow the Crowd: On QoE for Internet Applications Tobias Hoßfeld www3.informatik.uni-wuerzburg.dewww.t-hossfeld.de
Whatisthe Internet crowdconsuming? • Web and Cloud Applications • Online Video, Web Browsing, Downloads, Cloud Services, etc. • Why relevant? • Constitute dominant internet use cases • Generate relevant share of network traffic Global Consumer Internet Traffic Volume (Forecast).Source: Cisco VNI 2011.
Video Transmission over the Internet • UDP-based streaming • Unreliable transmission • Video quality affected • Artifacts may occur • Stimuli are visual degradationsor artifacts • HTTP streaming • Reliable transmission • Video quality not affected • But stalling may occur • Most stimuli/impairments are of temporal nature • YouTube uses HTTP streaming • Internet technology changes quality perception
Key Influence Factors on YouTube QoE • Interesting: no significantcorrelation of QoE and • initialdelay • video characteristics likeresolution, type of content,ratio of audio/video, etc. • users preference, whether they liked video • demographical features • Stalling frequency andstalling duration determinethe user perceived quality • Support vector machines and correlation coefficients
What is the influence of stalling on YouTube QoE? • Small number of interruptions strongly affect YouTube QoE • Provider (i.e. content and network provider) must avoid stalling • Total stalling time not sufficient for good QoE estimation • Monitoring of QoErequires sophisticated methods to capture stalling pattern, e.g. using DPI or directly at end user
Provider: Between the Devil and the Deep Blue Sea? • In case of insufficient resources, • „one“ has to choose between initial delays and stalling • What is worse for users? • Stalling has to be avoided,even at costs of initial delays • Current work: Is YouTube QoEmanagement beneficial for ISPs? • Users do „QoE management“ themselves – by pausing the video to prefetch contents and then to consume w/o interruptions • ISP may „invest“ in capacity, sophisticated traffic management, e.g. DASH and SVC • Exponential increase of costs wrt. quantile (of video corpus) • Delivering videos with about 120% of video bitrateas “rule of thumb”
Crowdsourcing • Crowdsourcing is a neologism composed of “crowd“ and “outsourcing“ literally, it means outsourcing to a (large, anonymous) crowd • All tasks are web-based micro jobs, typically little effort to fulfill • Crowdsourcing interesting for (QoE) user studies • large user panel, diversity of users, international users, • user studies can be executed in short time, • low costs in contrast to laboratory studies, • QoE tests for Internet applications with realistic settings Crowdsourcing is the act of taking a job traditionally performed by a designated agent (usually an employee) and outsourcing it to an undefined, generally large group of people in the form of an open call. Jeff Howe - Definition of Crowdsourcing “The White Paper Version“
Crowdsourcing Workflow 1 2 • Challenges due to remote setting • Unreliable QoEresults, no test moderator • Heterogeneousenvironment, devices, users 4 3 5
Countermeasures: Unreliability • Proper Test Design and statistical methods for filtering data • Consistency Tests • “Same” question is asked multiple times in different manner. • Example: user is asked about his origin country in the beginning and about his origin continent at the end. • Content Questions • Simple questions about the video clip, after watching the video. • Example, “Which sport was shown in the clip? A) Tennis. B) Soccer. C) Skiing.‘” • Application Usage Monitoring • Example: measuring the time the worker spends on the task • Example: monitor browser events and user reactions • Utilizefeaturesof crowdsourcing platform • Specializedcrowds, whichhavecertainskills, reliability, etc. • Conducttrainingsessions, two-stage tests • Payment accordingtoquality
Lessons Learned: Unreliable workers • FILTER LEVEL 1: • - wrong answers to content questions • different answers to the same questions • always selected same option • consistency questions: specified the wrong country/continent FILTER LEVEL 2: - did not notice stalling - perceived non-existent stalling • Many user ratings rejected • use simple testinstructions • avoid Java applets • takecareoflow Internet speed • avoid incentives for users to cheat, see Facebook results of student’s friends • User warning („Test not done carefully“) rejection rate decreased about 50% • improvements possible detailed analysis of (inter and intra-) rater reliability revealed: filtering too strict FILTERLEVEL 3: - did not watch all videos completely C1 C2 C3 C4 C5 C6 C7 Facebook First crowdsourcing tests
Crowdsourcing vs. Laboratory Studies • Crowdsourcing testswith Microworkers.com at Uni Würzburg • Lab studies within ACE 2.0 at FTW’s i:Lab • Similar results in laboratory and crowdsourcing study • Crowdsourcing appropriateforQoEtestsof Internet apps single stall event: 4 sec videoduration: 30 sec • 2,035 users from more than 60 countries participated in tests and rated 8,163 video. Payment was below 200,- Euro. • User diversity • Statisticallysignificantresults • Lowcosts, fast conduction
CrowdsourcingTestsfor HD Live Streaming • Live video streaming investigated via Microworkers and Facebook • Joint work within QUALINET STSM by Bruno Gardlo “Improving Reliability for Crowdsourcing-Based QoE Testing” • Strong differences due to worse viewing conditions and smaller screen resolutions context monitoring required, e.g. light conditions, • Critical, proper analysis of data, consider hidden influence factors
Qualinet “Crowdsourcing“ Task Force • Goal • Derive a methodologyand setup for crowdsourcing in QoE assessment, • Challenge crowdsourcing QoE assessment approach with usual “lab” methodologies, comparisonof QoE tests • Develop mechanisms and statistical approaches for identifying reliable ratings from remote crowdsourcing users, • Define requirements onto crowdsourcing platforms for improved QoE assessment. • Experiences with crowdsourcing • What are the main challenges? Reliability, environment/context monitoring, technical implementation, language problems … • Cartography for crowdsourcing use cases and mechanisms • Database with crowdsourcing results, e.g. impact of context factors on QoE, country, habits, … • Framework for crowdsourcing QoE tests • Results are implemented in framework “QualityCrowd” by TU Munich • Further information: https://www3.informatik.uni-wuerzburg.de/qoewiki/qualinet:crowd
Qualinet Task Force „Web andCloud Apps“ E-Mail Instant Messaging Customer Relationship Management Office Desktop Gaming • Technology changeandservicemigrationtocloudsstronglyimpactsuserperceptionandQoE • Currentactivites • DropboxQoE and mulicollaboration tools • QoE-aware adaptation mechanism for video streaming: DASH and SVC • Standardization: finalization of model and measurement methodology for web browsing QoE • https://www3.informatik.uni-wuerzburg.de/qoewiki/qualinet:webcloud Application Type Example OnLive Google Mail Facebook Chat SalesForce.com MS Office Live EyeOS Small High DegreeofInteractivity
Questions? www.t-hossfeld.dehossfeld@informatik.uni-wuerzburg.de