Search Results

Peer-to-Peer (P2P) and ad hoc networks have many points in common: both represent a decentralized self-organizing network structure. However few existing P2P algorithms are specifically designed to operate efficiently over ad hoc networks. And few ad hoc networks are designed to benefit from P2P infrastructures. We have worked on an epidemic dissemination protocol to maintain soft-state in a decentralized, peer-to-peer fashion, in ad hoc networks. This protocol is an enhancement of Passive Distributed Indexing (PDI) method proposed by Lindemann and Waldhorst. PDI is a method for distributing information in a P2P structure which is particularly suited to ad hoc networks, and does not involve an overlay topology. It makes use of broadcast messages to spread information via passive epidemic dissemination. We have enhanced PDI in order to reduce the number of broadcast messages when the search for an item may span several hops. Three enhancements are proposed: 1) Lazy query propagation to delay the propagation of query messages such that local responses can inhibit unnecessary search. 2) Quench waves to stop an already initiated query propagation when still possible. A decision algorithm determines whether to start a quench wave or not based solely on local information. 3) The use of Multi-Point Relay (MPR) or similar protocol and algorithm, to reduce redundant broadcast messages. This talk will present the current state of this research, and discuss several open aspects with the purpose of stimulating debate. The talk will also include an overview of related work such as epidemic models from biology, other epidemic protocols for P2P overlays and MANETs, including gossip (active) and promiscuous (passive) dissemination modes. Such protocols could be used for many different purposes, roughly any task requiring distributed soft-state maintenance in the ad hoc network, including DNS and identifier mappings, network monitoring and configuration, and so on. During the talk we will also exploit the possibility of using the protocol to disseminate service information for on-demand service deployment, and further, to assist in self-composing protocol structures.

Current studies on self-configuring and adaptive networks aim at developing specific and fixed protocols which are able to optimize their configuration in a variable network environment. In this talk we study the problem where the protocols need to cope with a defective execution, including the lossy execution or the injection of foreign code. One guiding question will be the creation of robust execution circuits which can distribute over a network and which continue their service despite parts of the implementation being knocked out. The ultimate goal is to enable protocol implementations to detect by themselves that they are malfunctioning and to let them correct their own operation mode and code base. As a show case, we present a protocol implementation which is robust against deletion (knock-out) of any single instruction, regardless whether this deletion affects the core protocol functionality or the resilience logic. The technique used in this first of its kind example is the self-modification of the running program, which can be naturally situated in an active networking context. Ultimately, a self-correcting protocol implementation has to constantly rewrite itself according to the (self-)observed performance. In this talk we will also point to related fields like self-correcting software, fault tolerant quantum computing and self-healing properties of biological systems. This is joint work with Lidia Yamamoto, Hitachi Europe.