The course aims to develop a thorough understanding of fault-tolerance in distributed systems. Due to their nature, distributed systems are inherently vulnerable to failures if not designed properly. At any time, a subset of the processes in a distributed system may fail by crashing or could be compromised and behave in a treacherous way (e.g., Byzantine failures). It is hence essential to design distributed systems and applications in such a way that they can adequately cope with failures. The lecture will present focus on how to deal with these issues.
By studying relevant methods and algorithms in details, the student will acquire a deep understanding of the issues at hand and the basic mechanisms to deal with such failures. Although the course will focus on the theory of such systems, it will also systematically draw links with practical applications, making it valuable to both theoreticians and practitioners.
Distributed algorithms, message-passing, synchrony models, agreement, replication, fault-tolerance, Byzantine agreement, self-stabilization, blockchain, randomized algorithms
✔ 専門力 | 教養力 | コミュニケーション力 | 展開力(探究力又は設定力) | 展開力(実践力又は解決力) |
Typical classes will alternate between slide-based presentations, interactive discussions, class exercises. Active contribution to class discussions is strongly encouraged.
授業計画 | 課題 | |
---|---|---|
第1回 | Introduction, models & definitions | Revision of basic concepts of distributed algorithms (models, synchrony, causality) |
第2回 | Synchronous consensus | 授業時に指示する. |
第3回 | Asynchronous consensus, FLP impossibility proof | 授業時に指示する. |
第4回 | Asynchronous consensus with unreliable failure detectors | 授業時に指示する. |
第5回 | Eventual leader election, Paxos | 授業時に指示する. |
第6回 | Byzantine consensus (I) | 授業時に指示する. |
第7回 | Byzantine consensus (II) | 授業時に指示する. |
第8回 | Randomized consensus | 授業時に指示する. |
第9回 | State-machine replication | 授業時に指示する. |
第10回 | Group membership, distributed transactions, atomic commit | 授業時に指示する. |
第11回 | Distributed ledger and blockchain mechanisms | 授業時に指示する. |
第12回 | Self-stabilization (I) | 授業時に指示する. |
第13回 | Self-stabilization (II) | 授業時に指示する. |
第14回 | Q&A + final test | 授業時に指示する. |
学修効果を上げるため,教科書や配布資料等の該当箇所を参照し,「毎授業」授業内容に関する予習と復習(課題含む)をそれぞれ概ね100分を目安に行うこと。
Course materials:
Slide copies, additional article copies, ...made available for download from the course webpage.
Reference Books:
1. Michel Raynal, "Fault-tolerant message-passing distributed systems," Springer, 2018. https://www.springer.com/gp/book/9783319941400
2. Ajay Kshemkalyani, Mukesh Singhal, "Distributed computing: principles, algorithms, and systems," Cambridge Uni. Press, 2011.
3. Wan Fokkink, "Distributed algorithms: an intuitive approach ," MIT Press, 2013.
4. Vijay K. Garg, "Elements of distributed computing," IEEE, 2002.
5. Gerard Tel, "Introduction to distributed algorithms (2nd ed.)," Cambridge Univ. Press, 2000.
6. Shlomi Dolev, "Self-Stabilization," MIT Press, 2000. https://mitpress.mit.edu/books/self-stabilization
Homework assignments and contribution to class discussion, assignments, reports (60%); and examination (40%).
Examination will assess the understanding of basic concepts of fault-tolerant distributed algorithms (problems, algorithms, and methodology) and reasoning (correctness and complexity).
Required knowledge:
Prior to taking this course, the student must have previously acquired,
through lectures or self-study, background knowledge on basic concepts
of fault-free distributed algorithms such as taught in the following
courses:
- CSC.T438 Distributed algorithm; __or__
- MCS.T406 (CSC.T406) Distributed Systems
Related course:
In the field of fault-tolerant and dependable computing systems, this course is complementary with:
- CSC.T524 Dependable Computing
- CSC.T438 Distributed Algorithms