Citation Context

...service. In deciding whether to implement multicast services at the IP layer or at end systems, there are two con icting considerations that we need to reconcile. According to the end-toend arguments =-=[17]-=-, a functionality should be (a) pushed to higher layers if possible, (b) unless implementing it at the lower layer can achieve large performance bene t that outweighs the cost of additional complexity...

Citation Context

...plain how Narada builds data delivery tree. Narada runs a distance vector protocol on top of the mesh. In order to avoid the well-known count-to-in nity problems, it employs a strategy similar to BGP =-=[16]-=-. Each member not only maintains the routing cost to every other member, but also maintains the path that leads to such a cost. Further, routing updates between neighbors contains both the cost to the...

Citation Context

...r layers if possible, (b) unless implementing it at the lower layer can achieve large performance bene t that outweighs the cost of additional complexity atthelower layer. In his seminal work in 1989 =-=[4]-=-, Deering argues that this second consideration should prevail and multicast should be implemented at the IP layer. This view so far has been widely accepted. IP Multicast is the rst signi cant featur...

Citation Context

...hods for construction of overlay spanning trees for data delivery. A rst approach is to construct the tree directly - that is, members explicitly select their parents from among the members they know =-=[8]-=-. Narada however constructs trees in a two-step process. First it constructs a richer connected graph that we termmesh. The mesh could in general be an arbitrary connected subgraph of the CVG (though ...

Citation Context

...inally, IP Multicast calls for changes at the infrastructural level, and this slows down the pace of deployment. While there have been attempts to partially address some of the issues at the IP layer =-=[9, 15, 21]-=-, fundamental concerns pertaining to the \stateful" architecture of IP Multicast and support for higher layer functionality haveremained unresolved. In this paper, we revisit the issue of whether mult...

Citation Context

... of links. Narada might dynamically add links to ensure connectivity of the virtual topology, and drop links it perceives as not useful. Self-con guration has been proposed in other contexts. AMRoute =-=[1]-=- allows for robust IP Multicast in mobile adhoc networks by exploiting user-multicast trees. Several reliable IP Multicast protocols [12, 13, 22] involve group members self-organizing into structures ...

Citation Context

...also need to consider the di cult case of abrupt failure. In such a case, failure should be detected locally and propagated to the rest of the group. In this paper, we assume a failstop failure model =-=[19]-=-, which means that once a member dies, it remains in that state, and the fact that the member is dead is detectable by other members. We explain the actions taken on member death with respect to Figur...

Citation Context

... other contexts. First, Narada distinguishes itself from normal routing protocols in that it changes the very topology over which routing is performed. Second, most existing self-con guring protocols =-=[12, 13, 14, 22]-=- assume native IP Multicast support. Narada attempts selfcon guration in the absence of a lower level multicast service, and this is fundamentally more challenging. We explain the distributed algorith...

Citation Context

...sumption signi cantly simpli es the partition recovery mechanisms in Gossamer, the members of the mesh could become partitioned from each other in event of failure of all rendezvous points. The MBone =-=[2]-=- and 6Bone [10] are popular existing examples of overlay networks. However, these are statically congured in a manual and adhoc fashion. Narada, on the other hand, strives for a self-con guring and e ...

Citation Context

... other contexts. First, Narada distinguishes itself from normal routing protocols in that it changes the very topology over which routing is performed. Second, most existing self-con guring protocols =-=[12, 13, 14, 22]-=- assume native IP Multicast support. Narada attempts selfcon guration in the absence of a lower level multicast service, and this is fundamentally more challenging. We explain the distributed algorith...

Citation Context

...inally, IP Multicast calls for changes at the infrastructural level, and this slows down the pace of deployment. While there have been attempts to partially address some of the issues at the IP layer =-=[9, 15, 21]-=-, fundamental concerns pertaining to the \stateful" architecture of IP Multicast and support for higher layer functionality haveremained unresolved. In this paper, we revisit the issue of whether mult...

Citation Context

... other contexts. First, Narada distinguishes itself from normal routing protocols in that it changes the very topology over which routing is performed. Second, most existing self-con guring protocols =-=[12, 13, 14, 22]-=- assume native IP Multicast support. Narada attempts selfcon guration in the absence of a lower level multicast service, and this is fundamentally more challenging. We explain the distributed algorith...

Citation Context

... other contexts. First, Narada distinguishes itself from normal routing protocols in that it changes the very topology over which routing is performed. Second, most existing self-con guring protocols =-=[12, 13, 14, 22]-=- assume native IP Multicast support. Narada attempts selfcon guration in the absence of a lower level multicast service, and this is fundamentally more challenging. We explain the distributed algorith...

Citation Context

...riodically generate a refresh message with monotonically increasing sequence number, which is disseminated along the mesh. Each member i keeps track of 1 An ideal mesh is a \Degree-Bounded K-spanner" =-=[11]-=- of the Complete Virtual Graph. The problem of constructing Degree-Bounded K-spanners of a graph has been widely studied in centralized settings that assume complete information and is NP-complete eve...

Citation Context

...medium. We also believe this is among the rst works that attempt to systematically evaluate the performance of a selforganizing overlay network protocol and the tradeo s in using overlay networks. In =-=[20]-=-, it was argued that the overlay approach is a fundamental technique to incrementally deploy services and evolve networks. We believe that the techniques and insights developed in this paper are gener...

Citation Context

...is was due to minor inaccuracies in delay measurements. However, in some cases, the reason was more fundamental and due to the policy based nature of Internet routing. This e ect has been reported in =-=[18]-=-. In the future we plan to conduct larger scale Internet experiments with emphasis on studying the dynamics of Narada, transient behavior and e ects of congestion and packet loss. 6. RELATED WORK The ...

Citation Context

...ces an edge between two points with a probability of e ;d L , where, d is the distance between vertices , L is the length of the longest possible edge and and are parameters. We use the Georgia Tech. =-=[23]-=- random graph generators to generate topologies of this model. Mapnet: Backbone connectivity and delay are modeled after actual ISP backbones that could span multiple continents. Connectivity informat...

Citation Context

...inally, IP Multicast calls for changes at the infrastructural level, and this slows down the pace of deployment. While there have been attempts to partially address some of the issues at the IP layer =-=[9, 15, 21]-=-, fundamental concerns pertaining to the \stateful" architecture of IP Multicast and support for higher layer functionality haveremained unresolved. In this paper, we revisit the issue of whether mult...

Citation Context

...congestion and packet loss. 6. RELATED WORK The works that come closest to the End System Multicast architecture we propose here and which share much ofour motivation are Yallcast [8] and Scattercast =-=[3]-=-. Both projects Cumulative Percentage of Pairs of Members 100 80 60 40 20 0 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 RDP Figure 17: Cumulative distribution of RDP challenge the appropriateness of using...

Citation Context

...ectivity. This information is collected by a route server from BGP routing tables of multiple geographically distributed routers with BGP connections to the server [6]. This data has been analyzed in =-=[5]-=- and has been shown to satisfy certain power laws. Random link delays of 8 ; 12 ms was assigned to each physical link. In our simulations, we usedbackbone topology sizes consisting of up to 1070 membe...