Назад в библиотеку

---

http://cgi.di.uoa.gr/...tue_b_3_liebeherr_talk_2slide.pdf

Two challenges for building large self-organizing overlay networks

Jorg Liebeherr

University of Virginia

Application-layer overlay networks, which organize sets of applications in virtual peer networks, have emerged as a new direction in networking research for deploying new network services. Since enhancements to the Internet core in the 1990s to provide QoS, multicast, or security, have either been slow, limited to special-purpose networks, or failed entirely, networking research now explores how the best-effort service of the Internet can be enhanced through application-layer overlay networks.

A fruitful area of research has been the search for solutions where a large number of peers can quickly (in the matter of seconds) self-organize in an overlay network without central control or coordination. Using topologies that are inspired by distributed hash tables, algorithms in computational geometries or other areas, research on large-scale overlay networks has shown that it is feasible to build self-organizing networks that grow to many thousand and, at least in principle, millions of peers.

The talk will address topics in overlay multicasting. Application-layer multicast has several attractive features:

(a) There is no requirement for multicast support in the layer-3 network;

(b) There is no need to allocate a global group identifier, such as an IP multicast address;

(c) Since data is sent via unicast, flow control, congestion control, and reliable delivery services that are available for unicast can be exploited. A drawback of application-layer multicast is that, since data is forwarded between end-systems, end-to- end latencies can be high. The talk will attempt to highlight two issues related to overlay networks:

* Economy-of-scale versus increased scalability: It appears that building overlay networks as logical graphs with a regular topology, that ignore the underlying network infrastructure, self-organize quickly, even if the number of peers is very large. However, such overlay networks can be poorly matched to the network-layer infrastructure, yielding high latencies and poor utilization of resources. Each overlay network solution trades-off economy-of-scale for increased scalability. Research on overlay topologies has only started to explore the design space of this trade- off.

* Programming Overlay Networks: Research on overlay networks has focused on the design of protocols to maintain and forward data in an overlay network, however, less attention has been given to the software development process of building application programs in such an environment. Clearly, the complexity of overlay network protocols calls for suitable application programming interfaces (APIs) and abstractions that do not require detailed knowledge of the overlay protocol, and, thereby, simplify the task of the application

programmer. A particular challenge is the development of programming paradigms that can evolve together with overlay network topologies.

Examples will be drawn from a currently ongoing project, called HyperCast, which builds an overlay network as a Delaunay triangulation graph, and which uses a construct, called overlay socket, that intends to simplify the task of overlay network programming.