dynamo five years later
play

Dynamo, Five Years Later Andy Gross Chief Architect, Basho - PowerPoint PPT Presentation

Aug 31, 2022 •144 likes •891 views

Dynamo, Five Years Later Andy Gross Chief Architect, Basho Technologies GOTO Chicago 2013 Tuesday, April 23, 13 Dynamo Published October 2007 @ SOSP Describes a collection of distributed systems techniques applied to low-latency key-value

  1. Dynamo, Five Years Later Andy Gross Chief Architect, Basho Technologies GOTO Chicago 2013 Tuesday, April 23, 13

  2. Dynamo Published October 2007 @ SOSP Describes a collection of distributed systems techniques applied to low-latency key-value storage Spawned (along with BigTable) many imitators, an industry (LinkedIn -> Voldemort, Facebook -> Cassandra) Authors nearly got fired from Amazon for publishing Tuesday, April 23, 13

  3. Riak - A Dynamo Clone First lines of first prototype written in Fall 2007 on a plane on the way to my Basho interview “Technical Debt” is another term we use at Basho for this code Mostly Erlang with some C/C++ Apache2 Licensed First release in 2009, 1.3 released 2/21/13 Tuesday, April 23, 13

  4. Basho Founded late 2007 by ex-Akamai people Currently ~120 employees, distributed, with offices in Cambridge, San Francisco, London, and Tokyo We sponsor of Riak Open Source We sell Riak Enterprise (Riak + Multi-DC replication) We just open sourced Riak CS (S3 clone backed by Riak Enterprise) Tuesday, April 23, 13

  5. Principles Always-writable Incrementally scalable Symmetrical Decentralized Heterogenous Focus on SLAs, tail latency Tuesday, April 23, 13

  6. Techniques Consistent Hashing Vector Clocks Read Repair Anti-Entropy Hinted Handoff Gossip Protocol Tuesday, April 23, 13

  7. Consistent Hashing Invented by Danny Lewin and others @ MIT/Akamai Minimizes remapping of keys when number of hash slots changes Originally applied to CDNs, used in Dynamo for replica placement Enables incremental scalability, even spread Minimizes hot spots Tuesday, April 23, 13

  8. Tuesday, April 23, 13

  9. Vector Clocks Introduced by Mattern et al, in 1988 Extends Lamport’s timestamps (1978) Each value in Dynamo tagged with vector clock Allows detection of stale values, logical siblings Tuesday, April 23, 13

  10. Read Repair Update stale versions opportunistically on reads (instead of writes) Pushes system toward consistency, after returning value to client Reflects focus on a cheap, always-available write path Tuesday, April 23, 13

  11. Hinted Handoff Any node can accept writes for other nodes if they’re down All messages include a destination Data accepted by node other than destination is handed off when node recovers As long as a single node is alive the cluster can accept a write Tuesday, April 23, 13

  12. Anti-Entropy Replicas maintain a Merkle Tree of keys and their versions/hashes Trees periodically exchanged with peer vnodes Merkle tree enables cheap comparison Only values with different hashes are exchanged Pushes system toward consistency Tuesday, April 23, 13

  13. Gossip Protocol Decentralized approach to managing global state Trades off atomicity of state changes for a decentralized approach Volume of gossip can overwhelm networks without care Tuesday, April 23, 13

  14. Hinted Handoff Tuesday, April 23, 13

  15. Hinted Handoff X • Node fails X X X X X X X Tuesday, April 23, 13

  16. Hinted Handoff X • Node fails X X • Requests go to fallback X X X X X hash(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA ”) Tuesday, April 23, 13

  17. Hinted Handoff • Node fails • Requests go to fallback • Node comes back hash(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA ”) Tuesday, April 23, 13

  18. Hinted Handoff • Node fails • Requests go to fallback • Node comes back • “Handoff” - data returns to recovered node hash(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA ”) Tuesday, April 23, 13

  19. Hinted Handoff • Node fails • Requests go to fallback • Node comes back • “Handoff” - data returns to recovered node • Normal operations resume hash(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA ”) Tuesday, April 23, 13

  20. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) Tuesday, April 23, 13

  21. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) client Riak Tuesday, April 23, 13

  22. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) client Riak Get Handler (FSM) Tuesday, April 23, 13

  23. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) client Riak hash(“blocks/ 6307C89A-710A-42CD-9FFB-2A6B39F983EA”) Get Handler (FSM) == 10, 11, 12 Tuesday, April 23, 13

  24. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) client Riak hash(“blocks/ 6307C89A-710A-42CD-9FFB-2A6B39F983EA”) Get Handler (FSM) == 10, 11, 12 Coordinating node Cluster 6 7 8 9 10 11 12 13 14 15 16 The Ring Tuesday, April 23, 13

  25. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) client Riak Get Handler (FSM) get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA ”) Coordinating node Cluster 6 7 8 9 10 11 12 13 14 15 16 The Ring Tuesday, April 23, 13

  26. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) client Riak Get Handler (FSM) R=2 Coordinating node Cluster 6 7 8 9 10 11 12 13 14 15 16 The Ring Tuesday, April 23, 13

  27. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) client Riak Get Handler (FSM) v1 R=2 Coordinating node Cluster 6 7 8 9 10 11 12 13 14 15 16 The Ring Tuesday, April 23, 13

  28. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) client Riak Get Handler (FSM) v1 v2 R=2 Tuesday, April 23, 13

  29. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) client v2 Riak Get Handler (FSM) v2 R=2 Tuesday, April 23, 13

  30. Anatomy of a Request get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA” ) v2 Tuesday, April 23, 13

  31. Read Repair get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA ”) client v2 Riak Get Handler (FSM) v1 v2 R=2 Coordinating node Cluster 6 7 8 9 10 11 12 13 14 15 16 v1 v2 Tuesday, April 23, 13

  32. Read Repair get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA ”) client v2 Riak Get Handler (FSM) v2 R=2 Coordinating node Cluster 6 7 8 9 10 11 12 13 14 15 16 v1 v2 Tuesday, April 23, 13

  33. Read Repair get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA ”) client v2 Riak Get Handler (FSM) v2 R=2 Coordinating node Cluster 6 7 8 9 10 11 12 13 14 15 16 v1 v1 v2 Tuesday, April 23, 13

  34. Read Repair get(“ blocks/6307C89A-710A-42CD-9FFB-2A6B39F983EA ”) client v2 Riak Get Handler (FSM) v2 R=2 Coordinating node Cluster v2 v2 6 7 8 9 10 11 12 13 14 15 16 v2 v2 v2 Tuesday, April 23, 13

  35. Riak Architecture Erlang/OTP Runtime Tuesday, April 23, 13

  36. Riak Architecture Erlang/OTP Runtime Riak KV Tuesday, April 23, 13

  37. Riak Architecture Erlang/OTP Runtime Client APIs Riak KV Tuesday, April 23, 13

  38. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Riak KV Tuesday, April 23, 13

  39. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Riak KV Tuesday, April 23, 13

  40. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Riak KV Tuesday, April 23, 13

  41. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Request Coordination Riak KV Tuesday, April 23, 13

  42. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Request Coordination get put delete map-reduce Riak KV Tuesday, April 23, 13

  43. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Request Coordination get put delete map-reduce Riak Core Riak KV Tuesday, April 23, 13

  44. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Request Coordination get put delete map-reduce consistent hashing Riak Core Riak KV Tuesday, April 23, 13

  45. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Request Coordination get put delete map-reduce consistent hashing Riak Core membership Riak KV Tuesday, April 23, 13

  46. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Request Coordination get put delete map-reduce consistent hashing handoff Riak Core membership Riak KV Tuesday, April 23, 13

  47. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Request Coordination get put delete map-reduce consistent hashing handoff Riak Core membership node-liveness Riak KV Tuesday, April 23, 13

  48. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Request Coordination get put delete map-reduce consistent hashing handoff gossip Riak Core membership node-liveness Riak KV Tuesday, April 23, 13

  49. Riak Architecture Erlang/OTP Runtime Client APIs HTTP Protocol Buffers Erlang local client Request Coordination get put delete map-reduce consistent hashing handoff gossip Riak Core membership node-liveness buckets Riak KV Tuesday, April 23, 13

Recommend

Dynamo, Five Years Later Andy Gross Chief Architect, Basho Technologies QCon London 2013

Dynamo, Five Years Later Andy Gross Chief Architect, Basho Technologies QCon London 2013

Dynamo, Five Years Later Andy Gross Chief Architect, Basho Technologies QCon London 2013 Friday, March 8, 13 Dynamo Published October 2007 @ SOSP Describes a collection of distributed systems techniques applied to low-latency key-value

931 views • 75 slides
Dynamo, Five Years Later Andy Gross Chief Architect, Basho Technologies QCon SF 2012 Thursday,

Dynamo, Five Years Later Andy Gross Chief Architect, Basho Technologies QCon SF 2012 Thursday,

Dynamo, Five Years Later Andy Gross Chief Architect, Basho Technologies QCon SF 2012 Thursday, November 8, 12 Dynamo Published October 2007 @ SOSP Describes a collection of distributed systems techniques applied to low-latency key-value

461 views • 33 slides
Dynamo Saurabh Agarwal What have we looked at so far ? Assumptions CAP Theorem SQL and

Dynamo Saurabh Agarwal What have we looked at so far ? Assumptions CAP Theorem SQL and

Dynamo Saurabh Agarwal What have we looked at so far ? Assumptions CAP Theorem SQL and NoSQL Hashing Origins of Dynamo This is year 2004 One Amazon was growing and other shrinking What led to Dynamo ? What led to Dynamo ?

425 views • 40 slides
DynamO Workshop Introduction to Event-Driven Dynamics and DynamO Dr Marcus N. Bannerman & Dr

DynamO Workshop Introduction to Event-Driven Dynamics and DynamO Dr Marcus N. Bannerman & Dr

DynamO Workshop Introduction to Event-Driven Dynamics and DynamO Dr Marcus N. Bannerman & Dr Leo Lue m.campbellbannerman@abdn.ac.uk leo.lue@strath.ac.uk MNB & LL DynamO Workshop 23/01/2015 1 / 32 Agenda Section Outline Agenda What

616 views • 32 slides
Dynamo Dynamo motivation Fast, available writes - Shopping cart: always enable purchases FLP:

Dynamo Dynamo motivation Fast, available writes - Shopping cart: always enable purchases FLP:

Dynamo Dynamo motivation Fast, available writes - Shopping cart: always enable purchases FLP: consistency and progress at odds - Paxos: must communicate with a quorum Performance: strict consistency = single copy - Updates serialized to

1.45k views • 110 slides
Dynamo Dynamo motivation Fast, available writes - Shopping cart: always enable purchases FLP:

Dynamo Dynamo motivation Fast, available writes - Shopping cart: always enable purchases FLP:

Dynamo Dynamo motivation Fast, available writes - Shopping cart: always enable purchases FLP: consistency and progress at odds - Paxos: must communicate with a quorum Strict consistency = single copy - Updates serialized to single copy

809 views • 52 slides
www.hdtsoccer.com All teams will wear the dynamo uniform with logos and the club has all rights

www.hdtsoccer.com All teams will wear the dynamo uniform with logos and the club has all rights

The competitive programs will be combined under the Houston Dynamo name. Recreational Programs will not be affected. Both VSC and Revolution will retain their existing recreational programs. www.hdtsoccer.com All teams will wear the dynamo

506 views • 8 slides
Dynamo & Bigtable CSCI 2270, Spring 2011 Irina Calciu Zikai Wang Dynamo Amazon's highly

Dynamo & Bigtable CSCI 2270, Spring 2011 Irina Calciu Zikai Wang Dynamo Amazon's highly

Dynamo & Bigtable CSCI 2270, Spring 2011 Irina Calciu Zikai Wang Dynamo Amazon's highly available key-value store Amazon's E-commerce Platform Hundreds of services (recommendations, order fulfillment, fraud detection, etc.) Millions of

461 views • 44 slides
Amazon Dynamo A Highly Available Key-value Store Present by Jian Fang jianf@cmu.edu What is

Amazon Dynamo A Highly Available Key-value Store Present by Jian Fang jianf@cmu.edu What is

Amazon Dynamo A Highly Available Key-value Store Present by Jian Fang jianf@cmu.edu What is Dynamo Eventually consistent key-value store Support scalable highly available data access Optimized for availability to maximize customer

479 views • 26 slides
Reliability at Scale A tale of Amazon Dynamo Presented by Yunhe Liu @ CS6410 Fall19 Slides

Reliability at Scale A tale of Amazon Dynamo Presented by Yunhe Liu @ CS6410 Fall19 Slides

Reliability at Scale A tale of Amazon Dynamo Presented by Yunhe Liu @ CS6410 Fall19 Slides referenced and borrowed from Max & Zhen P2P Systems: Storage [2017], VanHattum [2018] Dynamo: Amazons Highly Available Key-value Store

602 views • 43 slides
A Global Model Investigation of MJO Initiation for DYNAMO Guang Zhang Scripps Institution of

A Global Model Investigation of MJO Initiation for DYNAMO Guang Zhang Scripps Institution of

A Global Model Investigation of MJO Initiation for DYNAMO Guang Zhang Scripps Institution of Oceanography Subramanian and Zhang (2014, JGR) Objectives To investigate the MJO initiation in the Indian Ocean using the NCAR CAM3 and the DYNAMO

653 views • 21 slides
DynamO Workshop Tutorial: Thermostats, transport properties, and compression Dr Marcus N.

DynamO Workshop Tutorial: Thermostats, transport properties, and compression Dr Marcus N.

DynamO Workshop Tutorial: Thermostats, transport properties, and compression Dr Marcus N. Bannerman & Dr Leo Lue m.campbellbannerman@abdn.ac.uk leo.lue@strath.ac.uk MNB & LL DynamO Workshop 23/01/2015 1 / 11 This tutorial session

533 views • 11 slides
DYNAMO: DYnamic Inputs of Natural Conditions for Air Quality MOdels AQAST Year 3 Tiger Team

DYNAMO: DYnamic Inputs of Natural Conditions for Air Quality MOdels AQAST Year 3 Tiger Team

DYNAMO: DYnamic Inputs of Natural Conditions for Air Quality MOdels AQAST Year 3 Tiger Team Daniel Cohan, Loretta Mickley, Richard McNider, Arastoo Pour Biazar, Bryan Duncan Key Additional Participants in DYNAMO Air Quality Management

353 views • 14 slides
D r o u g h 33.6 -45.5 33.6 -45.5 11 years 5.4 years 11 years 5.4 years years t

D r o u g h 33.6 -45.5 33.6 -45.5 11 years 5.4 years 11 years 5.4 years years t

D r o u g h 33.6 -45.5 33.6 -45.5 11 years 5.4 years 11 years 5.4 years years t years Figure 2. Precipitation anomalies during 11 years of the dust bowl (1930-1940) Station Set-up Montana Mesonet January 2018 4 36 ranchers

533 views • 36 slides
Building a Dynamo Bridge between revit and excel Vdc Tdindustries Craig technology chappell

Building a Dynamo Bridge between revit and excel Vdc Tdindustries Craig technology chappell

Building a Dynamo Bridge between revit and excel Vdc Tdindustries Craig technology chappell specialist Main objectives Class Description This class will focus on using Dynamo to develop a bridge of information between Excel

292 views • 25 slides
DynamO Workshop Tutorial: Hard sphere simulation Dr Marcus N. Bannerman & Dr Leo Lue

DynamO Workshop Tutorial: Hard sphere simulation Dr Marcus N. Bannerman & Dr Leo Lue

DynamO Workshop Tutorial: Hard sphere simulation Dr Marcus N. Bannerman & Dr Leo Lue m.campbellbannerman@abdn.ac.uk leo.lue@strath.ac.uk MNB & LL DynamO Workshop 23/01/2015 1 / 9 Hard-Sphere fluids Section Outline Hard-Sphere

930 views • 9 slides
DYNAMO INVESTMENT PROJECTS 6 NOVEMBER 2014 AEB OFFICE Antonio Linares AEB Board Member,

DYNAMO INVESTMENT PROJECTS 6 NOVEMBER 2014 AEB OFFICE Antonio Linares AEB Board Member,

ASSOCIATION DYNAMO INVESTMENT PROJECTS 6 NOVEMBER 2014 AEB OFFICE Antonio Linares AEB Board Member, Construction Industry and Building Material Suppliers Committee Chairman ASSOCIATION DYNAMO INVESTMENT PROJECTS, 6 NOVEMBER 2014, MOSCOW

588 views • 35 slides
Dynamo: Theme and Varia0ons @shanley Riak 150

Dynamo: Theme and Varia0ons @shanley Riak 150

Dynamo: Theme and Varia0ons @shanley Riak 150 Services Global access Mul3ple machines

1.07k views • 93 slides
Introduction Need for a highly available Distributed Data Store During the holiday shopping

Introduction Need for a highly available Distributed Data Store During the holiday shopping

DYNAMO Amazons Highly Available Key-value Store PRESENTED BY GAURAV VAIDYA Some of the slides in this presentation have been taken from http://cs.nyu.edu/srg/talks/Dynamo.ppt Introduction Need for a highly available Distributed Data Store

1.02k views • 42 slides
Global 3D MHD Simulations of Dynamo Effects Lia Hankla University of Colorado, Boulder

Global 3D MHD Simulations of Dynamo Effects Lia Hankla University of Colorado, Boulder

Global 3D MHD Simulations of Dynamo Effects Lia Hankla University of Colorado, Boulder Bhupendra Mishra, Jake Simon, Mitch Begelman, Phil Armitage 1 Black Hole Binary State Transitions First discovered: Cygnus X-1 Usually: X-ray burst,

904 views • 12 slides
Predictability and Coupled Dynamics of MJO during DYNAMO: Role of diurnal SST in the initiation

Predictability and Coupled Dynamics of MJO during DYNAMO: Role of diurnal SST in the initiation

Predictability and Coupled Dynamics of MJO during DYNAMO: Role of diurnal SST in the initiation and intensity of the MJO2 Hyodae Seo Woods Hole Oceanographic Institution Art Miller, Aneesh Subramanian, Nick Cavanaugh Scripps Institution

397 views • 18 slides
Background Distributed Key/Value stores provide a simple put / get interface Great

Background Distributed Key/Value stores provide a simple put / get interface Great

Background Distributed Key/Value stores provide a simple put / get interface Great properties: scalability, availability, reliability Increasingly popular both within data centers Cassandra Dynamo Voldemort Dynamo: Amazon's Highly

844 views • 42 slides
Dynamo: Amazons Highly Available Key-value Store Giuseppe DeCandia, Deniz Hastorun, Madan

Dynamo: Amazons Highly Available Key-value Store Giuseppe DeCandia, Deniz Hastorun, Madan

Introduction Background Related Work System Architecture (core distributed systems techniques) Implementation Experiences & Lessons Learned Dynamo: Amazons Highly Available Key-value Store Giuseppe DeCandia, Deniz Hastorun, Madan

1.41k views • 111 slides
Data Storage Revolution Relational Databases Object Storage (put/get) Speed Dynamo

Data Storage Revolution Relational Databases Object Storage (put/get) Speed Dynamo

Data Storage Revolution Relational Databases Object Storage (put/get) Speed Dynamo Scalability PNUTS Availability CouchDB Throughput MemcacheDB No Complexity Cassandra Eventual Consistency Read Request Write

326 views • 28 slides

More recommend