Summary of Distributed Systems 2.3: System models

00:00:00 - 00:20:00

The different models for system behavior in distributed systems are discussed in this video. The three models are crash stop, crash recovery, and byzantine. The implications of each model are outlined, and it is shown how to account for potential problems in each case.

• 00:00:00 In this video, we see how to model systems in which nodes and networks might go wrong, by considering three aspects of a system model: networks, nodes, and timing. We discuss how reliable the communication link is assumed to be, as well as how long a message might take to be delivered. The third assumption is that the network is allowed to do anything, which is modeled with a malicious active adversary.
• 00:05:00 In this video, the three models of network behavior are introduced: a fair loss link, a reliable link, and a network partition. The model of network behavior for a crash stop failure is introduced: a process might crash and then be gone forever.
• 00:10:00 A synchronous system model assumes that messages arrive at a predetermined time, and that nodes always execute their code at a known speed. Another model, a partially synchronous model, allows for asynchronous behavior, in which messages may take a long time to arrive or nodes might be slow to execute.
• 00:15:00 This video discusses the different system models that are used in distributed systems. The three models are crash stop, crash recovery, and byzantine. The video outlines the implications of using each model and how to account for potential problems.
• 00:20:00 In this video, the behavior of a system is discussed. It is assumed that the system is fully trusted, with no nodes that can be "byzantine" or "crash stop." The timing of the system is also discussed, and it is assumed that it is asynchronous. However, if the system is partially synchronous, the guarantees of the distributed algorithm are lost.