March 2001 CBCB The Holy Grail: Media on Demand over Multicast Doron Rajwan CTO Bandwiz
CBCB Outline Problems with broadband content on demand The caching (CDN) solution What about multicast? Bandwiz solution: Broadband Content on Demand Over Multicast Summary
March 2001 CBCB The Challenge Broadband Content Delivery
CBCB Content Delivery Infrastructure
CBCB Unicast Content Delivery
CBCB We are now entering the Broadband Era Unicast Content Delivery
CBCB In the Broadband Era Unicast Content Delivery
CBCB In the Broadband Era Unicast Content Delivery
CBCB In the Broadband Era Unicast Content Delivery
CBCB In the Broadband Era Unicast Content Delivery
CBCB Current Situation Unicast Content Delivery
March 2001 CBCB Content Delivery Networks: Content network, caching, delivery from the edge
CBCB Moving Content to the Edge Directing requests to the Edge Content Delivery Networks
CBCB Moving Content to the Edge also moves the problem.. to the edge
CBCB Content Delivery Networks Again, a non scalable solution
CBCB CDN: Moving Content to the Edge Requires: Control, re-direction, update Overcomes: Network link congestion Network equipment congestion Distance delay Does NOT solve: Edge server congestion Storage problems Bandwidth utilization
March 2001 CBCB What About Multicast??
CBCB IP Multicast
CBCB
CBCB
CBCB
CBCB
CBCB
CBCB
CBCB
CBCB Multicast is Scalable, but… Does not support browsing / on-demand Unidirectional “non interactive” protocol Congestion Reliability Not widely deployed ISP’s worry that the network may crash Inter-domain connectivity issues Multiple, complicated, routing protocols
CBCB Bandwiz Solution Broadband Content Delivery On-Demand over Multicast
CBCB Coding for asynchronous (on demand) content delivery
CBCB Features Negligible receiver overhead (typically < 1%) Can be achieved with low computational complexity Encoder is flexible for ‘use optimization’ (file download or browsing) with a universal decoder Supports content “grouping” and differential decoding
CBCB But what about Media Streaming? Basic delivery scheme above requires receiving the entire file prior to display Streaming solution: Data can be decoded on the fly, enabling streaming-on-demand Streaming performance defined by: Bandwidth at the sender Peak bandwidth at the receiver Latency Computational complexity Latency is reduced exponentially with increased bandwidth at the server
CBCB Streaming On Demand Naïve solution – Carousel Today’s NVoD: N times bandwidth 1/N viewing latency Bandwiz Solution N times bandwidth 1/e^N viewing latency!!!
CBCB 90 minutes movie 200 kbps real time bit-rate total file size of 135 MB Transmitter bandwidth1 Mbps1.5 Mbps Receiver peak bandwidth1 Mbps320 Kbps CPU utilization~3%~1% Latency from ‘ click ’ to view50 seconds45 seconds Streaming - Example
CBCB Summary: Streaming On-Demand With ONLY 4 to 7 times the bandwidth of a single unicast transmission, and almost the same receiving bandwidth: Practically unlimited number of viewers Full screen quality viewing Most loss tolerant solution View from start when joining Short viewing latency (system parameter)
CBCB Servers/Bandwidth Savings in both Unicast/Multicast Today’s Unicast Bandwiz Unicast Bandwiz Multicast Total Output Bandwidth 1 Gbps 12 mbps # of Servers (Pentium PC) >1021 Viewing Latency 10 sec30 sec 2-3 full feature 100 min. MPEG movies 1 mbps per stream 1000 simultaneous viewers with broadband modems
CBCB Summary Multicast: Scalable, substantial resource savings However, unidirectional, non-interactive, same data to all receivers The Holy Grail: Use Multicast for On-Demand Delivery of Popular Content Streaming Solution: Optimal Tradeoff between Server Bandwidth and Delay
March 2001 CBCB THANK YOU