In Search of an Understandable Consensus Algorithm Diego Ongaro John Ousterhout

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

 In Search of an Understandable Consensus Algorithm 2014 USENIX Annual Technical Conference ongaro2014-raft ongaro2014-raft.txt, the project's staged copy 267af37a76990abfad71182403656be68989503f author-open computer-science

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

 0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them Quotation withheld because this record is not cleared for redistribution. Quotation withheld because this record is not cleared for redistribution. system-or-tool-building Quotation withheld because this record is not cleared for redistribution. design alternatives are chosen by asking two stated questions about UNDERSTANDABILITY - how hard is each to explain, and how easily can a reader completely grasp it and its implications - and the authors state plainly that this analysis is highly subjective Quotation withheld because this record is not cleared for redistribution. the first technique is problem decomposition: the problem is divided into pieces that can be solved, explained AND understood relatively independently, and the paper names the four it used Quotation withheld because this record is not cleared for redistribution. the second technique reduces the STATE SPACE rather than the code - fewer states to consider, more coherence, nondeterminism removed where possible - and it is made concrete by two rules about what logs are not allowed to do Quotation withheld because this record is not cleared for redistribution. NONDETERMINISM IS DELIBERATELY ADDED WHERE IT SHRINKS THE STATE SPACE: randomised approaches handle all possible choices in one fashion - choose any, it does not matter - so randomisation is used to simplify leader election even though the general policy was to eliminate it Quotation withheld because this record is not cleared for redistribution. the comparison is a within-subject crossover: every participant learns both algorithms and takes both quizzes, with the ORDER reversed for about half of them, so that individual differences and the experience gained from the first half are both accounted for Quotation withheld because this record is not cleared for redistribution. the two ways the experiment FAVOURED THE COMPETITOR are stated by the authors rather than left to a reader: some participants had prior experience with the rival algorithm, and the rival's teaching video was longer Quotation withheld because this record is not cleared for redistribution. the regression model's estimate of the quiz effect is reported as LARGER than the raw score difference, and the gap between the two is explained by prior experience with the rival algorithm rather than set aside Quotation withheld because this record is not cleared for redistribution. the participants' own preference is collected AND THEN DISCOUNTED by the authors, who name both reasons it may be unreliable: self-report against measured scores, and the participants' possible knowledge of the hypothesis being tested Quotation withheld because this record is not cleared for redistribution. the correctness argument is anchored to a formal specification that is itself the subject of the proof, and the paper states exactly how far the mechanisation reaches - one property mechanically proven, resting on invariants that have not been mechanically checked Quotation withheld because this record is not cleared for redistribution. what this limits: the liveness argument: Quotation withheld Quotation withheld because this record is not cleared for redistribution. what this limits: the safety proof as evidence: THIS PROOF RELIES ON INVARIANTS THAT HAVE NOT BEEN MECHANICALLY CHECKED (FOR EXAMPLE, WE HAVE NOT PROVEN THE TYPE SAFETY OF THE SPECIFICATION) Quotation withheld because this record is not cleared for redistribution. what this limits: the understandability claim, which is the paper's central selling point: PARTICIPANTS MAY HAVE BEEN BIASED BY KNOWLEDGE OF OUR HYPOTHESIS THAT RAFT IS EASIER TO UN[derstand], and the self-reports MAY BE LESS RELIABLE THAN ... QUIZ SCORES Quotation withheld because this record is not cleared for redistribution. what this limits: the performance comparison: Raft's strong-leader design rules out optimisations available to leaderless algorithms Quotation withheld because this record is not cleared for redistribution. what this limits: any performance comparison against optimised implementations: VARIOUS OPTIMIZATIONS HAVE BEEN PROPOSED IN THE LITERATURE FOR OTHER ALGORITHMS; MANY OF THESE COULD BE APPLIED TO RAFT, BUT WE LEAVE THIS TO FUTURE WORK Quotation withheld because this record is not cleared for redistribution. what this limits: the claim that this algorithm is more understandable by design: the alternatives were judged by how hard each is to explain and how easy it would be for a reader to understand, which are the authors' own assessments Quotation withheld because this record is not cleared for redistribution. what this limits: the finding that this algorithm is more understandable than Paxos: it was measured by an experimental study using upper-level undergraduate and graduate students in an Advanced Operating Systems course at Stanford and a distributed computing course elsewhere Quotation withheld because this record is not cleared for redistribution. what this limits: treating results about Paxos as results about this algorithm: the paper states it produces an equivalent result and is as efficient, but that its structure is different Quotation withheld because this record is not cleared for redistribution. what this limits: a description of this algorithm as deterministic by design: the authors state that although they tried in most cases to eliminate nondeterminism, in some situations nondeterminism actually improves understandability Quotation withheld because this record is not cleared for redistribution. eliminates: that the difference in quiz scores reflects individual differences in ability or learning carried over from whichever algorithm was studied first, rather than the algorithms' understandability · outcome: held Quotation withheld because this record is not cleared for redistribution. understandability Quotation withheld because this record is not cleared for redistribution. Paxos Quotation withheld because this record is not cleared for redistribution. decomposition Quotation withheld because this record is not cleared for redistribution. nondeterminism Quotation withheld because this record is not cleared for redistribution. state space Quotation withheld because this record is not cleared for redistribution. Quotation withheld because this record is not cleared for redistribution. Quotation withheld because this record is not cleared for redistribution. Quotation withheld because this record is not cleared for redistribution. Quotation withheld because this record is not cleared for redistribution. Quotation withheld because this record is not cleared for redistribution. Quotation withheld because this record is not cleared for redistribution. understandability state space strength of the relation: 1 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span gives state-space complexity as one example of how understandability was judged; operationalised-as is loose. Quotation withheld because this record is not cleared for redistribution. decomposition understandability status: the authors' DESIGN INTENT, stated as what they applied - not a measured outcome · direction on the measure: increases · the relation itself is separately evidenced · strength of the relation: 1 · the relation stated: decomposition is one of the techniques APPLIED TO IMPROVE understandability Quotation withheld because this record is not cleared for redistribution. Quotation withheld because this record is not cleared for redistribution. about this quote: QUOTE REPAIRED: the extractor split a hyphen-space and the quote was recorded faithfully with the damage in it, which made it unanchorable. The repaired string is an EXACT SUBSTRING of the staged text - that is why it was accepted, not because a rule produced it. The text as originally recorded is kept in evidence as extracted. Quotation withheld because this record is not cleared for redistribution. understandability Quotation withheld because this record is not cleared for redistribution. consensus algorithm about this quote: QUOTE REPAIRED: the extractor split a hyphen-space and the quote was recorded faithfully with the damage in it, which made it unanchorable. The repaired string is an EXACT SUBSTRING of the staged text - that is why it was accepted, not because a rule produced it. The text as originally recorded is kept in evidence as extracted. Quotation withheld because this record is not cleared for redistribution. replicated state machine Quotation withheld because this record is not cleared for redistribution. regression about this quote: EXTRACTION, NOT THE PAPER: the quote as extracted ended 'and s t n a p…', which is 'participants' reversed. It runs on into 'i f o r e b m u n' ('number of' reversed) - together the rotated y-axis label of Figure 15, which the extractor read bottom-to-top and placed inside this sentence. The sentence resumes 250 characters later at 'the order in…', so the true third factor is the learning order. The quote is TRIMMED rather than mended: joining the two halves would produce a string that reads correctly and exists in no document. Found by an internal tool; this was the only live quote in the corpus reaching into such a region. Quotation withheld because this record is not cleared for redistribution. statistical-significance about this quote: QUOTE REPAIRED: the extractor split a hyphen-space and the quote was recorded faithfully with the damage in it, which made it unanchorable. The repaired string is an EXACT SUBSTRING of the staged text - that is why it was accepted, not because a rule produced it. The text as originally recorded is kept in evidence as extracted. Quotation withheld because this record is not cleared for redistribution. t-test Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://raft.github.io/raft.pdf.