Considering Capacities in Application-Layer Multicast over Shared Access Networks

A huge number of application scenarios would greatly benefit from a group communication service in the Internet, but the existing IP Multicast solution still lacks global availability due to several technical and political issues [1]. As a consequence, Application-Layer Multicast (ALM) [2] arose as a promising solution to overcome this missing global multicast deployment. In ALM, forwarding functionality and membership state is exclusively handled on the participating end systems, following the well-known peer-to-peer paradigm. A typical ALM node has to handle higher traffic load than compared with native IP Multicast because packets are duplicated and forwarded on the nodes rather than in infrastructure routers. Therefore, ALM approaches were initially limited to stationary (comparably high-capacity) end systems that could cope with this higher load. However, recent and ongoing developments in the sectors of access technologies (e.g. VDSL, LTE etc.) and end system capabilities lead to scenarios in which employing ALM on mobile systems becomes feasible. Soon, these developments will render two-digit MBit upand downstream rates possible on every mobile device. This—together with alwayson possibilities at an affordable price— will lead to an intense growth in the number of users of such devices. Additionally, modern end system devices are able to communicate via more than one technology, potentially concurrently (compare Figure 1). Examples are cellular networks, WiFi, or even PANs like Bluetooth. We refer to these devices as multi-modal end systems, which we believe to hold great potential for ALM dissemination. At the same time, consumed contents will also grow in order to use the higher technical capabilities of the devices. Although individual capacities grow, some access technologies used are bound to upper capacity limits (e.g. in cellular network cells). Recently in the US, first service providers suffered from service outages triggered by too many iPhone users congesting the shared medium1. Although not clear if these congestions occured in the backbone or the access domain, the problem will be amplified in ALM scenarios where data is duplicated in end systems rather than infrastructure routers. Consider a videostreaming scenario, already being one of the most bandwidth-intense applications in the Internet. Since cellular networks will be the technology with the highest availability for mobile users, assume that a number of nodes in these access networks want to receive