In epidemic or gossip-based multicast protocols, each node simply relays each message to some random neighbors, such that all destinations receive it at least once with high probability. In sharp contrast, structured multicast protocols explicitly build and use a spanning tree to take advantage of efficient paths, and aim at having each message received exactly once. Unfortunately, when failures occur, the tree must be rebuilt. Gossiping thus provides simplicity and resilience at the expense of performance and resource efficiency. In this paper we propose a novel technique that exploits knowledge about the environment to schedule payload transmission when gossiping. The resulting protocol retains the desirable qualities of gossip, but approximates the performance of structured multicast. In some sense, instead of imposing structure by construction, we let it emerge from the operation of the gossip protocol. Experimental evaluation shows that this approach is effective even when knowledge about the environment is only approximate. 1

There is an inherent trade-off between epidemic and deterministic tree-based broadcast primitives. Tree-based approaches have a small message complexity in steady-state but are very fragile in the presence of faults. Gossip, or epidemic, protocols have a higher message complexity but also offer much higher resilience. This paper proposes an integrated broadcast scheme that combines both approaches. We use a low cost scheme to build and maintain broadcast trees embedded on a gossip-based overlay. The protocol sends the message payload preferably via tree branches but uses the remaining links of the gossip overlay for fast recovery and expedite tree healing. Experimental evaluation presented in the paper shows that our new strategy has a low overhead and that is able to support large number of faults while maintaining a high reliability. 1.

Recent decentralised event-based systems have focused on providing event delivery which scales with increasing number of processes. While the main focus of research has been on ensuring that processes maintain only a small amount of information on maintaining membership and routing, an important factor in achieving scalability for event-based peer-to-peer dissemination system is the number of events disseminated at the same time. This work presents a dynamic and fault tolerant cluster management method which can be used to coordinate concurrent access to resources in a peer-to-peer system. In the context of event-based dissemination systems the cluster management can be used to control the number of concurrently disseminated events. We present and analyse an algorithm implementing the proposed cluster management model in a faulttolerant and decentralised way. The algorithm provides for each cluster a limited set of tickets. A process which has obtained a ticket may send events corresponding to the resources of the cluster. The algorithm guarantees that no two processes ever issue an event corresponding to the same ticket at the same time. The cluster management model on its own has interesting properties which can be useful for many peer-to-peer applications.

Abstract. The scalability and resilience of epidemic multicast, also called probabilistic or gossip-based multicast, rests on its symmetry: Each participant node contributes the same share of bandwidth thus spreading the load and allowing for redundancy. On the other hand, the symmetry of gossiping means that it does not avoid nodes or links with less capacity. Unfortunately, one cannot naively avoid such symmetry without also endangering scalability and resilience. In this paper we point out how to break out of this dilemma, by lazily deferring message transmission according to a configurable policy. An experimental proof-of-concept illustrates the approach. 1

Within an effort for providing a layered architecture of services supporting multi-peer collaborative applications, this paper proposes a new type of consistency management aimed for applications where a large number of processes share a large set of replicated objects. Many such applications, like peer-to-peer collaborative environments for training or entertaining purposes, platforms for distributed monitoring and tuning of networks, rely on a fast propagation of updates on objects, however they also require a notion of consistent state update. To cope with these requirements and also ensure scalability, we propose the cluster consistency model.

During the past 30 years of the Internet revolution, the Internet has become a major force of change with an enormous effect on civilization. Consequently, computer networks have evolved into more complex system and become

Gossip, or epidemic, protocols have emerged as a powerful strategy to implement highly scalable and resilient reliable broadcast primitives on large scale peer-to-peer networks. Epidemic protocols are scalable because they distribute the load among all nodes in the system and resilient because they have an intrinsic level of redundancy that masks node and network failures. This chapter provides an introduction to gossip-based broadcast on largescale unstructured peer-to-peer overlay networks: it surveys the main results in the field, discusses techniques to build and maintain the overlays that support efficient dissemination strategies, and provides an in-depth discussion and experimental evaluation of two concrete protocols, named HyParView and Plumtree. The original publication is available at www.springerlink.com

One way to efficiently disseminate information in a P2P overlay is to rely on a spanning tree. However, in a tree, interior nodes support a much higher load than leaf nodes. Also, the failure of a single node can break the tree, impairing the reliability of the dissemination protocol. These problems can be addressed by using multiple trees, such that each node is interior in just a few trees and a leaf node in the remaining; the multiple trees approach allows to achieve load distribution and also to send redundant information for fault-tolerance. This paper proposes Thicket, a decentralized algorithm to efficiently build and maintain such multiple trees over a single unstructured overlay network. The algorithm has been implemented and is extensively evaluated using simulation in a P2P overlay with 10.000 nodes.

In this paper we propose a novel probabilistic broadcast protocol that reduces the average end-to-end latency by dynamically adapting to network topology and traffic conditions. It does so by using an unique strategy that consists in adjusting the fanout and preferred targets for different gossip rounds as a function of the properties of each node. Node classification is light-weight and integrated in the protocol membership management. Furthermore, each node is not required to have full knowledge of the group membership or of the network topology. The paper shows how the protocol can be configured and evaluates its performance with a detailed simulation model. 1