|
|
On-demand Media Distribution on Application-Layer Overlay (May. 2002 - Present) | ||||||
|
On-demand media distribution has gained intensive consideration recently due to its promising usage in a rich set of Internet-based services such as media distribution, distance education, video on demand, etc. The primary challenge here is to enable efficient, scalable and on-demand media delivery. By efficient, we mean that the system should make minimum use of media server and network resources. By scalable, we mean that the media distribution system should accommodate a large number of clients dispersed in the wide area. By on-demand, we mean that the client is free to access any portion of the media stream (e.g., fast forwarding and rewind in VoD) and the access latency must be bounded by a small delay time. An ideal media distribution system should meet all the three requirements. To achieve the goal of scalability and efficiency, a general approach is multicast. The nature of multicast is synchronous, i.e., data is sent to all receivers simultaneously. However, on-demand media distribution requires data to be delivered in an asynchronous manner, i.e., users want to receive data at different times. To apply multicast in the context of on-demand media distribution, different solutions have been proposed. The basic idea is to repeat the same content on multiple multicast channels over time. Asynchronous clients are either enforced to be synchronized at the price of service delays, or demanded to participate in several multicast sessions. Besides the above limitations, these solutions face another problem when deployed in the real world: they all assume the multicast support from underlying network. However, the deployment of IP multicast in the Internet has been slow and severely limited. Recently, application-layer multicast has been proposed to compensate for the absence of IP multicast. The main idea is to build an overlay network among end hosts. To enable multicast functionalities, end hosts cooperate with each other to relay data. Current research in this area has not considered applying application-layer multicast in the context of on-demand media distribution yet. Even if this approach is feasible, the intrinsic conflict between multicast and on-demand delivery still remain unresolved. In fact, the rise of application-layer overlay gives us a chance to avoid the above conflict by altering the nature of multicast, which is unachievable in the context of IP multicast. To this end, we propose a new mechanism, Asynchronous Multicast, to directly support on-demand data delivery. Operating on application-layer overlay network, asynchronous multicast relies on peer end hosts to relay data for each other. The key to asynchronous data transmission is cooperative buffering. By enabling data buffering on the relaying nodes, requests at different times can be satisfied by the same stream, achieving effcient asynchronous media delivery. While impossible in IP multicast since routers can buffer very limited amount of data, it can be realized on the application layer, where relaying nodes are end hosts or proxies with strong buffering capabilities. Asynchronous multicast is superior to the existing media distribution techniques in the following aspects:
|
IP Multicast
Application-Layer Multicast
Asynchronous Multicast | ||||||
| Our current analysis and performance evaluation have revealed that: First, the required server bandwidth of asynchronous multicast defeats the theoretical lower bound of traditional multicast-based solutions. Second, with respect to bandwidth consumption on the backbone network, the benefit introduced by asynchronous multicast overshadows the topological inefficiency of application overlay. As the next step, we plan to validate our solution via real-world network experimentation. | |||||||
Publications:
| |||||||