Análise de desempenho de soluções para armazenamento estável de dados.

Abstract

Fault tolerant applications often need to stabilize data in order to provide dependable computation. Ideas for services for data stabilizing abound, for example: synchronous operations on magnetic disks, data replication through the network, or even non-volatile RAM. This different ideas lead to different performance fingerprints. In this dissertation we focus on two problems: i) The stable storage problem, with architecture and implementations, and ii) The performance evaluation of this kind of service. Our work comprehends the study of the stable storage problem and solutions, study of methods for performance evaluation of this solutions, implementation of a network data-replication based solution and the performance evaluation of this solution with the classic ones, based on synchronous data write procedures. With the results, we assure the viability of our implementation and identify a series of trade offs in stable storage systems.