** AT&T, Telefonica, France Telecom, PCCW –Telcos that finally deployed IP multicast to provide IPTV. ** These IPTVs are quality-assured with fast channel switching – very similar to cable TVs.
** Log gathers all information about when a user switches a channel, and when the user leaves the channel.
This architecture shows IP backbone and the rest, access network. A multiplexer device, DSLAM, aggregates traffic from several hundreds of user' and connects to the high speed ISP network. Backbone uses static IP multicast tree. This means that TV head end streams all the live channels to each DSLAM at all times assuming there will be at least user watching each channel. Then, only one or two channels are delivered from DSLAM to the local user because of limited bandwidth. ***** guaranteed quality: End-to-end (from TV head end to the last mile).
Rewind function for past scenes are supported by VoD servers in by unicast, this is because rewind request is different for individual user.
Now based on the logs collected from a real IPTV network, I will highlight some of the most important viewing behaviors with their implications on design. *** This is the very first analysis!!!
The graph here shows … Average time spent on a channel is ** This is in high contrast to PPLive, where you have to wait for several minutes until you change a channel.
We saw very strong correlation both in time as well as in channels.
Channel popularity is highly skewed. The graph on the left shows.. The graph on the right shows…
** Not only there is temporal correlation within a day or a week, but also correlation within a program. ** We can make two interesting observations: 1) people join asynchronously over the program duration (need for rewind!), 2) constant churn of viewers (could be hard for P2P design)
Current Telco’s IPTV design is costly in that ** 1) static multicast delivers channels to DSLAMs even when no user is watching, and ** 2) rewind is unicast – incurring backbone crossing traffic and scalability concern at peak hours. As the future design, we propose Telco-managed P2P TV.
** 1) Not over PC, it is home gate or stb ** 2) Bandwidth is dedicated and symmetric – not like other P2P where ***** 3) Peer selection can be optimized to minimize the network cost (which is to reduce the backbone cost and contain traffic within DSLAM!), -- as opposed to what is now based on preventing free-riding.
Live TV: Replace static IP multicast by Telco managed P2P Rewind TV: Assist VoD servers to provide rewind function, where users jumps to the beginning of the ongoing program We can reply the real trace and compare network cost.
Static IP multicast tree – very simple design, but waste bandwidth a lot P2P TV – channel only arrive to DSLAM Difference between dynamic IP multicast and P2P TV: DSLAM cost will be roughly doubled because P2P traffic needs to go up and down, while backbone cost will stay the same.
We assume users want to watch from the beginning of the on-going channel. We mapped the program schedule with the channel switching logs to identify how much rewind a user wants to do on a given channel switching. High-correlation of channels and time.
On Next-Generation Telco-Managed P2P TV Architectures Meeyoung Cha MPI-SWS Pablo Rodriguez Telefonica Research Sue Moon KAIST Jon Crowcroft U of Cambridge
Motivation <ul><li>Nationwide IPTV deployment </li></ul><ul><li>Understand implications of user behavior </li></ul><ul><li>Study the impact of Telco-managed P2P IPTV system </li></ul>
<ul><li>IPTV Network Studied </li></ul><ul><li>One of the first </li></ul><ul><li>One of the largest </li></ul><ul><li>150 TV channels </li></ul><ul><li>250,000 subscribers </li></ul><ul><li>1-month detailed log </li></ul>
P2P Live TV <ul><li>P2P TV can carry a larger number of channels </li></ul>Why? Few hundred users per DSLAM Users not active all the time Focus skewed to few channels Static IP multicast (num channels * num DSLAMs) Lower bound of P2P TV average = 17% Total backbone traffic
P2P-Assisted Rewind TV <ul><li>P2P can support 95% of rewind traffic </li></ul>VoD server load without P2P (100%) VoD load with P2P (5%) P2P traffic served within DSLAM (80%)
Contributions <ul><li>1. The first detailed study on Telco’s IPTV </li></ul><ul><li>2. Understand implications of user behaviors </li></ul><ul><li>3. Evaluate Telco-manage P2P TV design </li></ul><ul><ul><li>Huge potential for live and rewind TV </li></ul></ul>