A Host-Based Multicast (HBM) Solution for Group Communications

Abstract

In this paper we argue that overlay multicast is an important technology for applications requiring a group communication service. With this approach end-hosts (running the application), dedicated servers and/or border routers automatically self-organize into a distribution topology where data is disseminated. This topology can be composed of both unicast connections and native multicast islands (e.g. within each site). Therefore it offers a group communication service to all hosts, even those located in a site that does not have access to native multicast routing. One of the issues raised is the set up of an efficient and robust overlay topology. In this paper we discuss several possible solutions. We show in particular the benefits of having a centralized approach, of using redundant links and updating the topology based on a host stability criteria.

On robustness in application-level multicast: the case of HBM

“…Basic Idea: The HBM protocol [7,6] automatically creates a virtual overlay topology, which by default is a shared tree, between the various group members (sources and receivers), using point-to-point UDP tunnels. Everything is under the control of a single host, the Rendez-vous Point (RP).…”

“…In order to reduce the probability of overlay topology partition in front of one or more node failures, we introduce Redundant Virtual Links (or RVL) [7] to the overlay topology created by HBM. Since the RP has a full knowledge of group members and creates/manages this topology, it can easily add a certain number of RVLs.…”

“…For instance with the first flavor no difference is made between leaves and transit nodes, whereas the remaining four flavors limit the number of RVLs that can be attached to a leaf. This distinction makes sense since nodes with limited processing or communication capabilities are always moved toward the leaves of the overlay shared tree topology [7]. Strategy I: any number of RVLs can be attached to any node.…”

“…They enable every host to participate in multicast sessions, no matter whether it has access to native multicast or not. The HBM protocol [7] is one such protocol. HBM is by nature centralized, everything being controlled by a single host, called Rendez-Vous Point (or RP).…”

Advanced Transition/Cluster Key Management Scheme for End-System Multicast Protocol

“…Each group member collects the metrics between itself and the other group members. The End-System Multicast (ESM) can be classified into two main categories: 1) tree first approaches [3]; 2) Mesh first approaches, like NARADA [4], centralized protocol as both ESM [5] and Host-based Multicast (HBM) [6], and semi centralized ESM as described in [1,7]. In [8], a highly scalable locating algorithm is proposed to gradually direct newcomers to a set of their closest nodes without inducing high overhead.…”