Presentation on theme: "People-level network analytics Monitor user workload characteristics Tailor (reprogram) network based on this Nishanth Sastry King’s College London."— Presentation transcript:
People-level network analytics Monitor user workload characteristics Tailor (reprogram) network based on this Nishanth Sastry King’s College London
Why this should be on EPSRC’s agenda? Why now [Timeliness ] – New closed loops allow people-level analytics IPTV, Catch-up TV, shortened URLs (bit.ly), Google analytics… Social media as URL launchers == “analytics plane” – Networks programmable as never before (SDN) Why UK [Novelty and ambition ] – strong in reprogramming of networks (all of session one!) – Strong in analytics/measurements (several in this room itself – you know who you are!) – Bring them together and make a difference!
People-Analytics Driven Network Designs Social network information for content placement – For timing push of news updates: TailGate (WWW’12) – Selective replication:Scellato et al (WWW’11), Buzztraq Easy to predict what you watch on Catch-up! – Speculative recording of people’s favourite programmes can halve BBC iPlayer network footprint (WWW’13) Adult video - people flexible on what to watch, so long as they have not seen it before (IMC’13) – Need to replace most watched vids for returning users! – Need to reconsider LRU policies in this setting
Research issues Integrating people-level info with flow-level programmability and virtualisation constructs – Granularity of “people-level” info different from SDN – Need new programming models/adapters Reliability of data/noisiness Heterogeneity (Heavy users/light users, different geographical densities) Extending to support cellular networks – Programmability is a completely different ball game Machine characteristics-driven M2M networks?