Systems where partial failure is the normal case.
13 topics · 41 curated works
Written on this field
Distributed systems is the rare engineering discipline whose canon is mostly negative results. Thirteen papers, and the comfortable assumption each one destroyed.
No prior grounding assumed.
Eventually Consistent - Revisited
Werner Vogels · 2008
Argues eventual consistency is a deliberate engineering trade-off large-scale systems make in favour of availability, not a lapse from a…
+1 more at this level
Assumes you know the vocabulary.
The Byzantine Generals Problem
Leslie Lamport, Robert Shostak & Marshall Pease · 1982
Proves that reaching agreement in the presence of arbitrarily faulty components requires more than two-thirds of participants to be honest.
+13 more at this level
Primary sources and full treatments.
Time, Clocks, and the Ordering of Events in a Distributed System
Leslie Lamport · 1978
Defines happened-before and logical clocks, giving distributed systems a usable notion of order without shared time.
+24 more at this level
12 of 41 works
Werner Vogels
Argues eventual consistency is a deliberate engineering trade-off large-scale systems make in favour of availability, not a lapse from a strong-consistency ideal they failed to reach.
link checked 17 Sept 2026Notes on this copyJames Lewis & Martin Fowler
Defines microservices by a cluster of common characteristics - independent deployability, decentralised data management - rather than a single rule, and argues the style trades simplicity for operational complexity.
link checked 17 Sept 2026Notes on this copyLeslie Lamport, Robert Shostak & Marshall Pease
Proves that reaching agreement in the presence of arbitrarily faulty components requires more than two-thirds of participants to be honest.
Assumes no distributed systems background — the problem is posed as a puzzle before any mathematics arrives
20 pageslink checked 17 Sept 2026Notes on this copyJim Gray
Analyses real outage data to argue that software bugs and operator mistakes, not hardware, are the dominant cause of system failure, reframing fault tolerance as a software and process problem rather than a hardware one.
link checked 17 Sept 2026Notes on this copyMaarten van Steen & Andrew S. Tanenbaum
Teaches distributed systems around a small set of recurring problems - naming, communication, coordination, consistency and fault tolerance - rather than as a tour of unrelated case studies.
600 pageslink checked 17 Sept 2026Notes on this copyLeslie Lamport
Restates Paxos stripped of the parliamentary parable, arguing the protocol itself is simple once separated from the proof that it works.
link checked 17 Sept 2026Notes on this copySanjay Ghemawat, Howard Gobioff & Shun-Tak Leung
Argues that component failure should be treated as the normal case rather than the exception, and designs a file system around relaxed consistency and chunk-level replication instead of POSIX semantics.
link checked 17 Sept 2026Notes on this copyJeffrey Dean & Sanjay Ghemawat
Shows that a huge class of data-processing jobs can be expressed as two functions, letting the runtime handle parallelism, failure and data placement.
link checked 17 Sept 2026Notes on this copyMike Burrows
Argues most distributed systems need a small, highly available lock and naming service rather than a full consensus library, and that hiding Paxos behind a simple lock API is what made it usable in practice.
link checked 17 Sept 2026Notes on this copyTushar D. Chandra, Robert Griesemer & Joshua Redstone
Argues the gap between the Paxos algorithm and a production consensus system is itself substantial engineering, covering leader election, disk corruption and group membership changes the original paper leaves unspecified.
link checked 17 Sept 2026Notes on this copyPat Helland
Argues large-scale systems should abandon ACID distributed transactions in favour of idempotent operations and business-level compensation, because coordination across machines does not scale.
link checked 17 Sept 2026Notes on this copyAvinash Lakshman & Prashant Malik
Combines Dynamo's partitioning and replication with Bigtable's column-family data model, giving up any single point of coordination entirely.
link checked 17 Sept 2026Notes on this copy