Use this URL to cite or link to this record in EThOS:
Title: Performing tasks and reaching agreement in distributed systems prone to adversarial crash-failures
Author: Mirek, Jarosław D.
ISNI:       0000 0004 8501 6609
Awarding Body: University of Liverpool
Current Institution: University of Liverpool
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
This dissertation tackles several questions in distributed computing and fault-tolerance. It consists of four main, more or less independent, chapters that all fit into a certain bigger picture - the problem of how distributed algorithms work, when they are executed in the presence of an adversary causing specific distractions. We study fault-tolerance for two fundamental problems in distributed computing: the Do-All and the Consensus problems. In the Do-All problem we expect t tasks to be performed by p processors in a reliable way - this means that we demand all the tasks to be performed, as a result of an algorithm execution, even though processors may be crashed by an adversary. The system is synchronized with a global clock and processors communicate via a multiple-access channel (MAC), which restricts simultaneous transmissions. We use work as a performance measure, that accrues the number of available processors steps throughout the execution of an algorithm. Within this framework, we analyze the Do-All problem on a MAC against different, novel adversarial scenarios. In particular, we define a class of Ordered adaptive adversaries, which cause crashes online according to some partial order of the participating processors, that is fixed by the adversary before the execution. Furthermore, we consider a class of Round-Delay adaptive adversaries, who see random choices of parties with a delay. In the second and third part of this thesis, we investigate the Do-All problem with extended assumptions about tasks. In the former part, we analyze the complexity of algorithms performing arbitrary length tasks, depending on whether preemption is available or not. By the means of preemption, we understand that performing tasks can be abandoned partway through without losing progress. The latter part consists of solutions for the Do-All problem with dependent tasks being in a partial order relation, where some tasks need to be performed before others. Finally, we consider the Consensus problem in which n processes must agree on a common value. We examine the problem in a synchronous message-passing model against Constrained adversaries, resembling Ordered adaptive adversaries. While commonly used Strongly-Adaptive adversaries model malicious attacks and Non-Adaptive ones - faults that occur independently of the parties executing the algorithm, Constrained adversaries model more realistic scenarios with fault-prone dependent processes, e.g., in hierarchical or dependent software/hardware systems. Results from the fourth part prove that there is a separation between adversarial scenarios typically considered in the literature.
Supervisor: Kowalski, Dariusz ; Wong, Prudence ; Klonowski, Marek Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral