Absolute Slicing in Peer-to-peer Systems

Abstract

Peer-to-peer (P2P) systems are slowly moving from application-specific architectures to a generic service- oriented design framework. The idea is to allow a dynamic collection of P2P applications to cohabit into a single system, with applications starting and terminating at will, or even changing their requirements at run-time. This raises an interesting problem in connection with managing resource assignment in a large-scale, heterogeneous and unreliable environment. Recently, the distributed slicing service has been proposed to allow for an automatic partitioning of P2P networks into groups (slices) that represent a controllable amount of some resource. A particular instantiation of such service has been described, called ordered slicing, in which nodes are ranked based on some metrics and then assigned to a slice based on their position in the ranking. In this paper, we present an alternative version of the problem called absolute slicing. Here, the goal is to assign a specified number of nodes to a slice and maintain such assignment in spite of churn. We propose a simple algorithm that solves the problem by combining well-known protocols such as peer sampling and aggregation, and we experimentally evaluate its performance.