consensus   1236

« earlier    

In search of a simple consensus algorithm
In this post: (1) covered an availability limitation of the Raft protocol (2) demonstrated that modern implementations of Raft are subject to it (3) described an existing simpler approach to the problem of consensus (4) showed that its toy 500-lines implementation has performance similar to Etcd but doesn't suffer from Raft's performance penalty
consensus  paxos  availability  actors 
6 days ago by mpm
The Writings of Leslie Lamport
This document contains descriptions of almost all my technical papers and electronic versions of many of them for downloading.
algorithm  distributed  research  reliability  time  consensus 
11 days ago by brandon.w.barry
Tendermint - Blockchain Consensus
Byzantine fault-tolerant replicated state machines in any programming language
blockchain  consensus  protocol  decentralized  crypto 
11 days ago by fifigyuri
Spanner vs. Calvin
I found it very difficult to find cases where an ideal implementation of Spanner theoretically outperforms an ideal implementation of Calvin.
consistency  scaling  consensus  database  performance 
17 days ago by mpm
I'm a Quaker, and we could tell a lot of stories about both the triumphs and tragedies of seeking unity as a body. One of the things that makes it work better in a Quaker setting than I've experienced in other places (like collectively-run feminist and lesbian bodies back in the 80s and 90s) is that there is a strong shared value placed on the process—which is to say, we really deeply believe that it is possible to find a shared perspective and that how we get to a decision is at least as important as the decision itself. So, for instance, there is a Quaker practice of minuting when unity can't be found—we accept that sometimes it's just not possible. It's rare, but some of the most powerful Quaker business meetings I've been at have ended in a painful shared recognition that we couldn't all come together on an issue.

Also, we do this unity shit all the fucking time. It's how we make every decision in our monthly meetings (the odd Quaker name for the body that meets every week for worship and meets for business meeting once a month, hence the name); in committee meetings; at Quarterly and Yearly meetings. Our kids start having business meetings "in the manner of Friends" as, like, kindergartners. As a result we have vast experience in every kind of outcome: someone strongly disagrees and blocks a decision; someone strongly disagrees and, realizing that they are out of step with the body as a whole, "steps aside"; easy unity; and hard unity that comes only after weeks, months, or years of working on an issue. We have meetings called "threshing meetings" where we get together over a difficult decision with no intention of making a decision, but only to listen to each other. We are deeply trained not to speak more than once on an issue unless we are very strongly led to do so; we allow silence between speakers; we have no taboo against changing one's mind—it is expected that listening deeply to the insights and experiences of others will change hearts and minds, and when someone says, "I came into this meeting feeling strongly that X, but after listening to you all, I am now comfortable proceeding with Y," we accept that as part of the process. We rarely dig in our heels or defend an initial position (if someone is regularly doing this, they are not in order and will usually be called out on it).

In addition, every meeting is clerked. It has one person (sometimes two sharing the role) whose job is to facilitate things, but more deeply to listen and try to hear an emerging unity. You'll hear a clerk say, "I'd like to propose a minute," and then they will sum up what we call the "sense of the meeting," and surprisingly often, even when we didn't think we were coming together, they will have found the place where we agree. If a meeting is getting contentious, the clerk will often call for silence. Hands are up all over the room, everybody wants to throw in their two cents, and the clerk will say, "Let's settling into some silent worship," and after a couple of minutes, when you can feel the energy of the room has come down a bit, they'll say, "OK. Now, I'd like to hear from people who haven't spoken before, or who have points that haven't been made before, or who feel very strongly about being heard." Usually 9 out of 10 of those hands won't go back up, because on reflection a person has cooled off, or realized they were just repeating someone else.

We also say, "This Friend speaks my mind" when we strongly agree with someone. That way you don't have to stand up and spend five minutes repeating what that person said.

A clear can ask someone if they're willing to stand aside, if it's clear they're alone outside unity. This is why we talk about the "sense of the meeting" and not "universal agreement." If someone does step aside, this is minuted. When my meeting was thinking of building a meetinghouse after decades of meeting in rented spaces, one woman who had belonged to a meeting with a meetinghouse really opposed the idea. She felt, based on her experience on the Building and Grounds committee there, that a meeting that owns a building can find its energy and purpose diverted to supporting the building. Finally, when it had become clear that the meeting as a whole was ready to buy property, she stood up and said, "You all know I oppose this, and that hasn't changed. I do feel, however, that my concerns have been heard, and I recognize that I am out of step. I'd like to stand aside on this decision." This meant that we didn't have to persuade her to support building a meetinghouse; she thought we'd listened and taken her concerns seriously, and she was willing to let it go.

We take clerking really seriously. We study how to do it; there are workshops and books and retreats on how to do it well. Everybody clerks at some point. Every committee has a clerk; even a casual one-off meeting to discuss, oh, what we're serving at the annual picnic will start by choosing someone to clerk the meeting. In this way we build experience, and we also discover who has particular gifts at clerking. A good clerk is a joy to behold; a weak one is a burden to be borne until their term ends. It is common in Quaker organizations to limit terms of service on committees and as clerks, so as to give everyone experience and also to avoid individuals becoming entrenched. I have a friend who has been holding a monthly meeting for worship for a particular purpose in his home every months for, oh, twenty years? He keeps it going; he hosts it; he provides food and tea every month. But he doesn't clerk it. At the beginning of every month's meeting, somebody volunteers.

Of course, we are also very aware that every contentious decision eventually gets made in part because some people leave. Same-sex marriage, for instance, has been supported by many liberal Quaker meetings for decades. When my partner and I, then a same-sex couple, attended our first meeting at our local Friends meeting back in 1994 or so, somebody came up to us after the meeting to tell us that they did same-sex union ceremonies. They'd minuted their support not long before, and hadn't had a chance to actually do one yet. (The meeting has done quite a few by now; but not one for me and my partner, although we are still together.)

Every meeting that minuted support for same-sex marriage lost people over it. Sometimes unity is reached in part because the people who are not unity go away, leaving behind a bunch of people who agree with each other. Some of the more conservative branches of Quakerism are experiencing schisms, or at risk of them, over same-sex marriage. Heck, over whether it's OK to be gay at all. This is very painful, but hard to avoid.

We've also seen meetings go to hell over the most minor issues: the color of the new carpet, say. A friend of mine recently visited a meeting while traveling, and reported to me with an eyeroll that, in this time of political strife and the deep need for people to stand up for human rights and help the vulnerable, they were embroiled in a lengthy and contentious conflict over whether to install air conditioning in their meeting house.

We also can end up wasting a lot of time on trivial issues. Until you've sat in a poorly-clerked meeting of 50 people, all having their say on where the commas should go in a minute or epistle, while the first truly beautiful spring day of the year fades into twilight outside the window, you haven't really experienced Quakerism. At its best, Quaker process is transcendent, holy, a powerful force for human connection and good in the world. At its worst, you spend the best years of your life getting way too invested in whether the windows in the social hall have blinds or curtains.

I talked about a person not in unity standing aside. Although it's rare, Quaker process also allows for a meeting that is otherwise unified to make a decision even if Friend Decision-Blocker won't stand aside. I think I've seen this happen once in all my years of Quakering; it's a less-than-ideal outcome and always feels a bit like we've failed. But there are times when time is tight or action has to be taken quickly, and at those times "sense of the meeting" doesn't necessarily mean "everybody agrees 100%."

Yours in the Light, as we Quakers say,

Friend Orlop.
quaker  consensus 
24 days ago by kejadlen
Multileader WAN Paxos: Ruling the Archipelago with Fast Consensus
We present WPaxos, a multileader wide area network (WAN) Paxos protocol, that achieves low-latency high-throughput consensus across WAN deployments. WPaxos dynamically partitions the global object-space across multiple concurrent leaders that are deployed strategically using flexible quorums. This partitioning and emphasis on local operations allow our protocol to significantly outperform leaderless approaches, such as EPaxos, while maintaining the same consistency guarantees. Unlike statically partitioned multiple Paxos deployments, WPaxos adapts dynamically to the changing access locality through adaptive object stealing. The ability to quickly react to changing access locality not only speeds up the protocol, but also enables support for mini-transactions
paxos  scaling  consensus 
26 days ago by mpm
RT : "For a balanced panel, we would need 96 more Dr. " -. What looks like…
climate  consensus  from twitter
4 weeks ago by mattsaler

« earlier    

related tags

*  090516  0916  1984  academic  academic_paper  acm  activism  actors  agreement  algorithm  algorithms  architecture  article  availability  aws  bellyfeel  best-of  bitcoin  blockchain  books  bullshit  business  byzantine_generals  capitalism  captheorem  casper  centralisation  change-social  climate  climatechange  cluster  cmv  cockroachdb  cognitive_bias  comments  communication  community  computerscience  consensusreality  consistency  consortium  contract  coordination  corruption  critical_theory  crypto  cs  cultural_critique  cybersecurity  database  databases  decentralized  decision  decision_making  decisions  definition  delicious  democracy  democrats  deniers  design  dev  development  dewey  directors  discourse  discussion  disgust  disney  dissent  distributed-computing  distributed-system  distributed-systems  distributed  distributed_systems  distributedsystems  dns  dogma  downloaded  duckspeak  echochamber  economics  election  emotionalism  ethereum  exceptionalism  experience  fail-fast  feminism  filesystem  filetype:pdf  financial  forcedmemes  github  globalwarming  go  golang  goodthink  gop  group_conversation  groupthink  gryadka  habermas  hairy_arms  herd  hierarchy  history  hlc  hm  howto  ideology  illiberalism  implementation  important  indeterminacy  internet  internet_slang  james_william  journalism  joycamp  kafka  kv  lang:fr  language  latency  leader  lecture  ledger  liberal  logging  lying  making  management  meeting  meetings  metafilter  model  monitoring  morality  morallicensing  n  nietzsche  normative_theory  norms  nosql  nousistanselection  online_deliberation  open_source  opensource  opinion_dynamics  orchestration  ordering  organisation  organizing  paper  papers  paxos  peirce  performance  pitfalls  polder  political-science  political_philosophy  politics  pos  post-truth  power  pragmatism  presentation  private  process  programming  protocol  protocols  psychology  public  public_opinion  public_reason  public_sphere  python  quaker  raft  rationality  reading  reddit  reliability  research  review-article  rhetoric  richardlindzen  scalability  scale  scaling  scapegoating  scientific_literature  security  services  shard  slavespeak  slides  smart  social_theory  socialmedia  sociocracy  software  soma  sql  sqlite  startup  statism  stories  street  system  threatnarrative  throughput  time  toread  validation  vcs  victimhood  visualisation  visualization  zookeeper 

Copy this bookmark: