英语论文网

Data Management for Peer-to-Peer Computing: A Vision1
Philip A. Bernstein2, Fausto Giunchiglia3, Anastasios Kementsietsidis4,
John Mylopoulos3, Luciano Serafini5, and Ilya Zaihrayeu2
Abstract. We motivate special database problems introduced by peer-to-peer computing and propose the LocalRelational Model (LRM) to solve some of them. As well代写留学生论文, we summarize a formalization of LRM, present anarchitecture for a prototype implementation, and discuss open research questions.
1. Introduction
Peer-to-peer (hereafter P2P) computing consists of an open-ended network of distributed computational peers,where each peer can exchange data and services with a set of other peers, called acquaintances. Peers are fully
autonomous in choosing their acquaintances. Moreover, we assume that there is no global control in the form ofa global registry, global services, or global resource management, nor a global schema or data repository.Systems such as Napster and Gnutella popularized the P2P paradigm as a version of distributed computing lyingbetween traditional distributed systems and the web. The former is rich in services but requires considerableoverhead to launch and has a relatively static, controlled architecture. The latter is a dynamic, anyone-to-anyone
architecture with little startup costs but limited services. By contrast, P2P offers an evolving architecture wherepeers come and go, choose whom they deal with, and enjoy some traditional distributed services with less startupcost.
We are interested in data management issues raised by this paradigm, where each peer may have data to sharewith other peers. For simplicity, we assume that each peer’s database is relational. Since the data residing indifferent databases may have semantic inter-dependencies, we allow peers to specify coordination formulas thatexplain how the data in one peer must relate to data in an acquaintance. For example, the patient database of afamily doctor and that of a pharmacy may want to coordinate their information about a particular patient, theprescriptions she has received, and the dates when these prescriptions were filled. Coordination may meansomething as simple as propagating all updates to the Prescription and Medication relations, assumed to exist inboth databases. In addition, we'd like a query expressed with respect to one database to be able to use relevant
databases at acquaintances, acquaintances of those acquaintances, and so on. To accomplish this, we expect theP2P data management system to use coordination formulas for recursively decomposing the query into subqueriesthat are evaluated with respect to the databases of acquaintances. Coordination formulas may also act assoft constraints or guide the propagation of updates. In addition, peers need an acquaintance initializationprotocol where two peers exchange views of their respective databases and agree on levels of coordination
between them. The level of coordination should be dynamic, in the sense that acquaintances may start with littlecoordination, strengthen it over time with more coordination formulas, and eventually abandon it when tasks andinterests change.In such a dynamic setting, we cannot assume the existence of a global schema for all databases in a P2P network,
or even those of all acquainted databases. Moreover, peers should be able to establish and evolve acquaintances,preferably with little human intervention. Thus, we need to avoid protracted tasks by skilled data