Serviços de processamento tolerantes a faltas para sistemas distribuídos assíncronos.

Abstract

Providing fault tolerance mechanisms through services can decrease the complexity in developing dependable distributed applications. This is because the application programmer will be able to use the services without needing to know how these services were implemented . In this work we propose hardware and software fault-tolerant processing services for off-the-shelf distributed systems. In these systems there no upper bound for the message passing and communication delays, so, they are asynchronous systems. Further, we study the dependability degree that can be achieved using these services and present some implementation strategies. Finally, we defined a protocol for message ordering which is required for managing redundancy into the group of replicated processors over which the processing services are built. Dependable applications are, in a lesser or greater extent, critical. This fact yields the necessity of validating the implementation of all the services being used by these applications. In order to facilitate this task, a number of system models has been proposed in the literature, each one having its own advantages and disadvantages. In the case of the processing services being proposed in this work, we followed a hybrid approach that gathers the facilities of well know system models, the asynchronous system model with unreliable failure detectors and the timed asynchronous system model, eliminating their respective inconveniences. The objective is to obtain an extended timed asynchronous system model that allows the specification of simple and practical solutions. Such a model incorporates two extra services: an unreliable failure detection service and a reliable broadcast service. The protocols that implement these services are defined and validated, assuring the characteristics of the referred extended model.