Mind the Gap: Modelling Video Delivery Under Expected Periods of Disconnection Argyrios G. Tasiopoulos, Ioannis Psaras, and George Pavlou Department of Electronic & Electrical Engineering University College London ACM CHANTS 2014 Maui-Hawaii
Outline Introduction – Motivation – Aims – Scope Model Evaluation Conclusions & Future work ACM CHANTS 2014 Maui-Hawaii
Motivation In big cities (London, New York, Tokyo, etc.) public transport is the preferred mean of travelling Almost all commuters are equipped with smartphones Live streaming is an increasingly popular smartphone application ACM CHANTS 2014 Maui-Hawaii INTRODUCTIO N
A Usual Situation ACM CHANTS 2014 Maui-Hawaii INTRODUCTIO N
Aims Questions that we want to answer: – Could we quantify commuters’ Quality of Experience (QoE) of video streaming? – Could we find the benefit of cooperative video streaming in terms of QoE in this setting? Yes!!! If we could create a model able to calculate over time for each case the: – Playback time – Playback disruptions time ACM CHANTS 2014 Maui-Hawaii INTRODUCTIO N
Scope We focus on the “Tube” environment – Expected Intermittent Connectivity Internet connectivity only available in train stations We focus on “Wi-Fi” Internet connectivity – Offered usually for “free” – Trend of installing “Wi-Fi” hotspots ACM CHANTS 2014 Maui-Hawaii INTRODUCTIO N
Outline Introduction Model – Fundamentals – Utility Functions Evaluation Conclusions & Future work ACM CHANTS 2014 Maui-Hawaii
Fundamentals Expected Intermittent Connectivity: – We know the time that a train spends in a station Connection Period: for station i – We also know the time needed to reach the next one Disconnection Period: – An epoch i,,consists of a connection and disconnection period ACM CHANTS 2014 Maui-Hawaii MODEL
Intermittent Connectivity ACM CHANTS 2014 Maui-Hawaii MODEL Time t=1 ii+1i+1
Fundamentals (2/4) Limited bandwidth of Wi-Fi AP – Limited by Wi-Fi AP and network infrastructure – We assume that is shared equally among users Video/Content consists of chunks – Collection of video packets – Specific bit-rate – Specific playback duration Two video streaming approaches, a basic and a cooperative one ACM CHANTS 2014 Maui-Hawaii MODEL
Fundamentals (3/4) ACM CHANTS 2014 Maui-Hawaii MODEL For the basic video streaming approach we use the “Pull” characterization – Since users retrieve a video individually by “pulling” it chunk by chunk The chunks received over an epoch i are: The time is discrete since protocols need some time to reallocate the bandwidth
Fundamentals (4/4) Next we name the cooperative video streaming approach as “PUSH” – Since the commuters have to “Pull” some chunks and then they have to “Share” them with the rest of their group, of magnitude at moment Thus, the number of chunks received over an epoch i for this approach are: ACM CHANTS 2014 Maui-Hawaii MODEL
Utility Functions Playback time until epoch i: Playback disruption time until epoch i: “Pull” utility function: – Where is the delay sensitivity coefficient “Push” utility function: – Where is the energy spent by a user for broadcasting his/her content in terms of playback time that could be downloaded from a WiFi AP MODEL ACM CHANTS 2014 Maui-Hawaii
Outline Introduction Model Evaluation – Theoretical Setting – Realistic Setting Conclusions & Future work ACM CHANTS 2014 Maui-Hawaii
Theoretical Setting In this setting: – All epochs have the same overall duration – The disconnection to duration ratio,, is constant for all epochs: – The number of users remain stable over all epochs The following results produced for: – 100 commuters, shared bandwidth 54 Mbps, chunk bit-rate 419 Kbps and playback duration 5’’, delay sensitivity 1, and a connection period of 30’’ EVALUATION ACM CHANTS 2014 Maui-Hawaii
Theoretical Results Pull Case EVALUATION ACM CHANTS 2014 Maui-Hawaii
Theoretical Results PUSH Case EVALUATION ACM CHANTS 2014 Maui-Hawaii Group size: 10 commuters
Pull-PUSH Comparison M=1 EVALUATION ACM CHANTS 2014 Maui-Hawaii
Realistic Setting In this setting: – Here we use real traces of London Underground – We focus on Victoria Line which has the least intersections But still… we have to find a way in order to form groups of users interested in the same content EVALUATION ACM CHANTS 2014 Maui-Hawaii
Content Assignment Algorithm Each user who does not watch a content in the beginning of an epoch creates a content with probability Else he/she joins an existed content, created during this epoch, according to Zipf’s distribution with Zipf’s exponent Please note that the in these content there is always included the “empty” one EVALUATION ACM CHANTS 2014 Maui-Hawaii
Realistic Set.: PUSH results over time EVALUATION ACM CHANTS 2014 Maui-Hawaii
Realistic Set.: Pull results over time EVALUATION ACM CHANTS 2014 Maui-Hawaii
Realistic Set.: PUSH results over epochs for various “p_new” EVALUATION ACM CHANTS 2014 Maui-Hawaii
Outline Introduction Model Evaluation Conclusions & Future work ACM CHANTS 2014 Maui-Hawaii
Conclusions and Future Work Conclusions: – We quantified the QoE of video streaming for the “Tube” setting – We found the gains offered by the collaboration among users in case of “few” available contents Future work: – Include the cellular case – Provide incentives for users to form a groups Maybe we will crease the content diversity ACM CHANTS 2014 Maui-Hawaii CONCLUSIONS & FUTURE WORK
Thank You