Distributed eventual leader election in the crash-recovery and general omission failure models.

Abstract

Distributed applications are present in many aspects of everyday life. Banking, healthcare or transportation are examples of such applications. These applications are built on top of distributed systems. Roughly speaking, a distributed system is composed of a set of processes that collaborate among them to achieve a common goal. When building such systems, designers have to cope with several issues, such as different synchrony assumptions and failure occurrence. Distributed systems must ensure that the delivered service is trustworthy.Agreement problems compose a fundamental class of problems in distributed systems. All agreement problems follow the same pattern: all processes must agree on some common decision. Most of the agreement problems can be considered as a particular instance of the Consensus problem. Hence, they can be solved by reduction to consensus. However, a fundamental impossibility result, namely (FLP), states that in an asynchronous distributed system it is impossible to achieve consensus deterministically when at least one process may fail. A way to circumvent this obstacle is by using unreliable failure detectors. A failure detector allows to encapsulate synchrony assumptions of the system, providing (possibly incorrect) information about process failures. A particular failure detector, called Omega, has been shown to be the weakest failure detector for solving consensus with a majority of correct processes. Informally, Omega lies on providing an eventual leader election mechanism.