Технологии поиска данных в одноранговых сетях, построенных на основе распределенных хеш-таблиц

Abstract

Проведен анализ архитектуры одноранговых сетей, построенных на основе распределенных хэш-таблиц, а также введены метрики для сравнения алгоритмов поиска с целью определения оптимального.

Здійснено аналіз архітектури однорангових мереж, побудованих на основі розподілених хеш-таблиць, та введено метрики для порівняння алгоритмів пошуку з метою визначення оптимального.

Introduction. The necessity of the large data sets processing, increasing the scalability and fault tolerance lead to the development of peer-to-peer networks, that require, unlike client-server architecture, the use of complex algorithms for data search. Thus, it is necessary to determine a peer-to-peer network with the optimal search algorithm according to the certain criteria. Purpose. The purpose of this paper is to analyze the constructed peer-to-peer networks based on distributed hash tables and to determine the optimal network according to some metrics. Methods. To determine the optimal system we have introduced the set of metrics, which correspond to the set of criteria’s which are critical to the systems comparison. The set of criteria’s is as follows: the number of steps in the search algorithm, the operation of calculating the «distance» between the elements of the network, and practical implementation of the system. Next, the criteria are ranked by importance and to each criterion there is a certain coefficient (weight) which is assigned so that the more important criterion strongly influences on the result than the less important. The linear convolution has been calculated for each system and the best system is the one which has the minimal result. Results. According to the result of calculations, the optimal search criteria is proved by Kademlia system, primarily due to the unique metric of distance between elements of the network. Conclusion. Analysis and comparison of peer-to-peer networks constructed on the basis of distributed hash tables, shows that, despite the fact that each of them is able to solve the problem of scalability and fault tolerance, the optimum system is Kademlia.