Multicast Video over IP

Conventional networking fails to achieve efficiency in the distribution of content. Point-to-point protocols are not scalable and are unable to broadcast video simultaneously to large numbers of users.

However, a system for Scalable Media Delivery (SMD) has been created to allow video to be multicast over several different channels to a large number of users. Clients "tune in" to videos being broadcast. Unlike point-to-point protocols, the video server will have little trouble being able to handle the addition of large numbers of new users tuning in.

Scalable Media Delivery (SMD) uses a powerful set of multicast protocols to reduce network bottlenecks and smooth data flow. Instead of fighting to get the same bits, SMD promotes communcal cooperation through the use of local caches. Every client is also a server that can draw upon an array of distributed network resources. Thus, clients tuning into video can access the video from any number of nearby clients rather than a single server. In addition, error correction is handled locally rather than by a central server which handles all errors throughout the network.

Conventional Networks:

Use Point-to-Point communication.

Deal with EVERY user as a unique individual.

Does not scale.

Geographically unaware.

Costly error correction.

A First Step: MBone

To address the problems of point-to-point communications, the network research community came up with MBone. Characteristics of MBone include:

Send one broadcast for all to listen.

Reduces redundant bandwidth.

Does not offer efficient time shifting.

Costly error correction.

Broadcast server solely responsible for repairs, and can easily become inundated with NACKS.

Scalable Media Delivery

Developed at the MIT Media Lab by Roger Kermode, Scalable Media Delivery (SMD) addresses a number of the shortcomings of MBone. Research Papers are available that detail the protocol.

The Benefits of Scalable Media Delivery Include:

Uses distributed caches to store broadcast info

Every client has the potential to be a server.

Encourages cooperation between networked machinesby enabling them to repair their peers.

Reduces network bottlenecks.

Can be deployed over EXISTING network infrastructure

Allows for distributed time shift re-broadcasts.

Introduces a system of distributed fault tolerance.

The future of networking

As companies become increasingly connected to their customers over the internet scaleable networking will become a cornerstone of smooth business operation.

Conventional ISPs, or some intermediary will have to offer scaleable solutions that ease the burden of delivery from their commercial accounts.
As CATV companies expand their data services, distributed caches will become an increasingly dominant part of the consumer network landscape.
- Philips is working on a prototype that offers terabytes of storage (STORE IT). Possibly for a settop box?

Research Papers

R.G. Kermode, "Smart Network Caches: Localized Content and Application Negotiated Recovery Mechanisms for Multicast Media Distribution", Doctoral Dissertation, MIT Media Lab, June 1998. postscript (2423063 bytes), gzipped postscript (587654 bytes) pdf (1205322 bytes), gzipped pdf (656538 bytes) Abstract
R.G. Kermode, "Scoped Hybrid Automatic Repeat reQuest with Forward Error Correction (SHARQFEC)," To appear in ACM Sigcomm 98, September 1998, Vancouver, Canada. gzipped postscript (143524 bytes), gzipped pdf (126548 bytes)
Preliminary results are available for viewing.
J. Gemmell, E. Schooler, and R. Kermode, "An Architecture for Multicast Telepresentations", To appear: Journal of Computing and Information Technology.
J. Gemmell, E. Schooler, and R. Kermode, "A Scalable Multicast Architecture for One-to-many Telepresentations", IEEE Multimedia Systems 98, June 28 - July 1, 1998, Austin TX, USA

[ Scalable Media Deliver | Digital Life | Media Lab Home ]

Maintained by: Chris Metcalfe
Questions, comments? Email: multicast@media.mit.edu.