Um serviço de comunicação síncrono para nodos com semântica de falha controlada utilizando redes assíncronas.

Abstract

Replicated processing on independent processors is a common way to achieve faulttolerant processing. It is the basis for the reliable processing service offered by the Seljuk-Amoeba operating environment. In order to assure that the replicated processes will achieve order and agreement on the input messages received by the replicas, replicated processing normally uses atomic broadcast protocols. These protocols assume that communication between any two operational processes is synchronous, i.e. there is a known finite time bound for message transmission between any two processes that are executing in accordance with their specification. In this work, we study the dificulties involved in the development of a synchronous communication service implemented in software, built on top of conventional asynchronous networks. In order to overcome these dificulties, we propose a simulation of a TDMA (Time Division Multiple Access) access method to the communication media, by which a process can reserve a portion of a TDMA slot exclusively for its use. The proposal also details the scheduling of the tasks in the operating system, necessary to assure a time limit not only for the transmission delay itself, but also to the end-to-end transmission delay for the message ordering protocols under Seljuk-Amoeba, accounting for the transmission request at the sending side until the proper delivery of the message at the destination. Besides, we also suggest a flow control mechanism, in order to guarantee that the process' transmission requests are compatible with the reservation made on the TDMA slot.