Presentation on theme: "Can Internet Video-On-Demand be Profitable? Jiwon Park July 11,2012."— Presentation transcript:
Can Internet Video-On-Demand be Profitable? Jiwon Park July 11,2012
Authors Cheng Huang, Jin Li Microsoft Research Redmond, WA 98052 ACM SIGCOMM 2007 Association for Computing Machinery Categories and Subject Descriptors [Computer-Communication Networks]: Distributed Systems Keith W. Ross Polytechnic University Brooklyn, NY 11201
4 Background VoD(Video-on-demand) in the Internet has become an immensely popular service in recent years. But due to its high bandwidth requirements &popularity, it is also a costly service to provide. Using a nine-month trace from a client-server VoD deployment for MSN Video, we assess what the 95 percentile server bandwidth costs would have been if a peer-assisted employment had been instead used. Considering the design and potential benefits of peer-assisted video-on- demand, in which participating peers assist the server in delivering VoD content. The assistance is done in such a way that it provides the same user quality experience as pure client-server distribution. Motivation
4 Goals: Focusing on the single-video approach, whereby a peer only redistributes a video that it is currently watching. Showing that peer-assistance can dramatically reduce server bandwidth costs, particularly if peers prefetch content when there is spare upload capacity in the system. Developing a simple analytical model which captures many of the critical features of peer-assisted VoD, including its operational modes. Also considering the impact of peer-assisted VoD on the cross-traffic among ISPs. I f care is taken to localize the P2P traffic within the ISPs, we can eliminate the ISP cross traffic while still achieving important reductions in server bandwidth. 5
How Internet traffic will look in 5 years time 6 Internet traffic will quadruple between 2009 and 2014 online video will be the biggest driver of that growth. 91% of all consumer internet traffic in 2014 will be online video, which includes video watched in web browsers and Internet VOD. P2P traffic growth seems to be levelling out while online video is continuing to explode. We really are visual creatures, no doubt about it… (Source: Cisco’s Annual Visual Networking Index Forecast)Visual Networking Index
Current situation - None of the Internet VoD providers earn significant revenues from their services. Revenue model But given the enormous costs associated with client-server distribution due both to the increasing video quality & to the enormous demand the revenues very possibly will not cover the cost. Although VoD has become one of the most popular Internet services today, the service is, and will likely continue to be, unprofitable with client-server distribution. 7 Threat of current VoD system Embedded video advertisements SubscriptionsPay-per-views
Design approaches to peer-assisted VoD Server stores videos and guarantees that users playback the video at the playback rate without any quality degradation. Since peer assisted VoD can move a significant fraction of the uploading from the server to the peers, it can potentially dramatically reduce the publisher’s bandwidth costs. Single video approach : a peer only redistributes the video it is currently watching; not have watched and stored in the past. ex) A torrent in BitTorrent in which all peers in the torrent share exactly one file. Multiple video approach :a peer can redistribute a video that it previously viewed but not currently viewing. 8
Mathematical model for peer-assisted VoD Purpose: Futher investigate the potential benefits of peer-assisted VoD. Determine the aggregate upload resourses of the participating peers & compare it with the aggregate user demand. Single on demand video of rate r bps. Classify users according to their upload link bandwidths. Users arrive at the system in a Poisson process with parameter λ. The average upload bandwidth of an arriving user : μ=∑P m ω m Steady state Demand: D=r∑ρ m = r λ σ Supply: S=∑ω m ρ m =μλσ 9 mnumber of user types. ΩmΩm upload link bandwidth of type m user PmPm probability of type m user’s arrival. P m λUser arrival model(1≤m≤M)
Different ways redistribute vidio via peer to peer 1.No prefetching Download content at the playback rate(r). Not prefetch context for future needs. 2, Prefetching An important question arises in how to allocate the instantaneous surplus upload capacity among the peers in the system 10 No prefetching Prefetching Water-leveling policy Greedy policy
No prefetching policy Order these n users so that user n is the most recent to arrive. Thus user 1 has been in the system the longest. uj : upload bandwidth of the jth user user j is of type m with probability pm, so p(uj = wm) = pm. Let the state of the system : (u1, u2, · · ·, un) the rate required from the server : s(u1, u2, · · ·, un). 11
Prefetching models Should we devote all the surplus capacity to one peer, rapidly building a reservoir for that peer while neglecting the other peers? Or should we try to equally allocate the surplus bandwidth, building small reservoirs of content at each of the peers? 12 Water-leveling policy, which aims to equalize the reservoir levels of prefetched content across all the peers. Greedy policy, where each user simply dedicates its remaining upload bandwidth to the next user right after itself.. the remaining bandwidth at each user is recorded. allocates as much bandwidth as possible to the subsequent user.
P2P Methodologies Users arrive with poison distribution Exhaustive search for available upload BW 100 Video rate: 60 60 3040 010 100 0 0 70 Total Demand 60 x 4 = 240 Total Support 100+40+30+100 = 270
System status If>If Support > Demand –Surplus mode, small server load If
"name": "System status If>If Support > Demand –Surplus mode, small server load If
Simulation: Non-early-departure Cbservations: a peer assisted distribution system had been used instead of the client-server system… 1.the server rate would have been dramatically reduced.! 2.at the current quality level, typically no server resources are needed. When the number of concurrent users is small, there is greater (normalized) variance in the upload capacity, so that peer-assisted VoD is more likely to run into a temporary deficit states that require server participation. 3. surplus mode due to the relatively low bitrates of the videos. we can easily offer much higher streaming quality and still trim the server rate significantly. 4. peer-assistance can be beneficial for both flash crowd (gold stream) and long-lasting (silver stream) videos.
Simulation: Early departure When video length > 30mins, 80%+ users don’t finish the whole video
Simulation: Full How to deal with buffer holes –As user may skip part of the video Two strategies –Conservative: Assume that user BW=0 after the first interaction –Optimistic: Ignore all interactions
ISP-unfriendly P2P VoD ISPs, based on business relations, will form economic entities –Traffic do not pass through the boundary won’t be charged ISP-unfriendly P2P will cause large amount of traffic.
Simulation results of friendly P2P Peers lies in different economic entities do not assist each other
CONCLUSION Consider the design and potential benefits of peerassisted video-on-demand. Using the nine-month MSN Video trace, we report on key observations of the characteristics from such a large scale VoD service. A theory is presented With peer-assistance and prefetching, we show the enormous potential cost savings to content providers. also examine the costs that peer-assisted VoD might place on local ISPs and explore how these costs can be minimized. Pros This paper gives a representative trace analysis about upload BW problems. Successfully address the importance of the P2P cross-ISP problem. Cons Weak and unrealistic P2P models. (so theortic) Unclear comparisons between each P2P strategies and simulations