|
|
|
|
|
|
|
Researchers developing active and programmable networks often face questions about the benefits such networks offer. It can be (and has been) argued that a flexible, robust, and programmable platform offers benefits to service providers and protocol developers by reducing the cost of deploying new services. However, examples of particular {\em application-level} benefits have been less than abundant. Moreover, built-in network support for specific applications is considered to violate the end-to-end argument. It thus seems that programmability offers the most benefit in a network service that is application-independent in its overall design, but can be customized to be useful to a large class of applications. In this talk, we will present {\em concast}, an Internet service that fits this description and also has a number of characteristics we believe are necessary for widespread deployment and success of new network services.
Concast is a many-to-one communication service that provides scalability through {\em abstraction} in much the same way as Internet multicast: a single address represents a {\em group} of participants. In the case of multicast, the participants are receivers; for concast they are senders. With multicast, the network keeps track of receivers and duplicates packets on the way. With concast, the network merges packets on the way to the (single) receiver. However, to provide benefits to a large class of applications, the service needs to be programmable, allowing applications to define the semantics of ``merging'' messages. Concast is interesting because it is a programmable service that is useful in solving real-world problems. The service is backward compatible with existing Internet protocols, can be incrementally deployed, provides significant benefits with only partial deployment, and has a number of other beneficial properties. We will present results from a simulation study and working prototype showing that concast provides significant benefits for applications such as layered-video and audio-mixing.
|
|
|
|
|
|
|
|