What We Talk About When We Talk About Distributed Systems

Transcription of this talk with list of resources:

http://videlalvaro.github.io/2015/12/learning-about-distributed-systems.html

Distributed Systems are a complex topic. There's abundant research about it but sometimes it is hard for a beginner to know where to start. I would like to outline the main concepts of distributed systems, so the interested person can have a clear path on how to start their own research as well.

In this talk I will review the different models: asynchronous vs. synchronous distributed systems; message passing vs shared memory communication; failure detectors and leader election problems; consensus and different kinds of replication.

I will also review a series of books on distributed systems in order to recommend the best one according to the topics we would like to learn about, or the problems we would like to solve.

The goal of the talk is to set a good foundation for people interested in learning more about distributed systems.

Transcript

1. WHAT WE TALK ABOUT
   WHEN WE TALK ABOUT
   DISTRIBUTED
   SYSTEMS
   ALVARO VIDELA - RABBITMQ
   

   View Slide

2. DISTRIBUTED
   SYSTEMS FOR
   THE IKEA FAMILY
   ALVARO VIDELA - RABBITMQ
   

   View Slide

3. DISTRIBUTED
   SYSTEMS
   

   View Slide

4. “A DISTRIBUTED SYSTEM IS
   ONE IN WHICH THE FAILURE
   OF A COMPUTER YOU DID NOT
   EVEN KNOW EXISTED CAN
   RENDER YOUR OWN
   COMPUTER UNUSABLE”
   Leslie Lamport
   

   View Slide

5. View Slide

6. View Slide

7. Google: define jargon
   

   View Slide

8. DISTRIBUTED
   SYSTEMS
   

   View Slide

9. DISTRIBUTED
   SYSTEMS
   • Many entities trying to solve a problem
   (nodes, processes)
   

   View Slide

10. DISTRIBUTED
    SYSTEMS
    • Many entities trying to solve a problem
    (nodes, processes)
    • Partial Knowledge
    

    View Slide

11. DISTRIBUTED
    SYSTEMS
    • Many entities trying to solve a problem
    (nodes, processes)
    • Partial Knowledge
    • Uncertainty
    

    View Slide

12. DEEP RABBIT
    HOLE
    

    View Slide

13. WHAT TO
    READ?
    

    View Slide

14. WHICH
    PAPERS?
    

    View Slide

15. View Slide

16. View Slide

17. View Slide

18. View Slide

19. View Slide

20. WHICH BOOKS?
    

    View Slide

21. View Slide

22. WHY?
    

    View Slide

23. http://tobielangel.com
    

    View Slide

24. THE PROBLEM
    

    View Slide

25. DIFFERENT MODELS
    

    View Slide

26. DIFFERENT MODELS
    • Timing Model
    

    View Slide

27. DIFFERENT MODELS
    • Timing Model
    • Inter Process Communication Used (IPC
    method)
    

    View Slide

28. DIFFERENT MODELS
    • Timing Model
    • Inter Process Communication Used (IPC
    method)
    • Failure Modes
    

    View Slide

29. TIMING MODEL
    

    View Slide

30. TIMING MODEL
    • Synchronous Model
    

    View Slide

31. TIMING MODEL
    • Synchronous Model
    • Asynchronous Model
    

    View Slide

32. TIMING MODEL
    • Synchronous Model
    • Asynchronous Model
    • Semi-synchronous Model
    

    View Slide

33. INTERPROCESS
    COMMUNICATION
    

    View Slide

34. INTERPROCESS
    COMMUNICATION
    • Message Passing
    

    View Slide

35. INTERPROCESS
    COMMUNICATION
    • Message Passing
    • Shared Memory
    

    View Slide

36. FAILURE MODES
    

    View Slide

37. FAILURE MODES
    • Crash-stop
    

    View Slide

38. FAILURE MODES
    • Crash-stop
    • Crash-recovery
    

    View Slide

39. FAILURE MODES
    • Crash-stop
    • Crash-recovery
    • Omission Faults
    

    View Slide

40. FAILURE MODES
    • Crash-stop
    • Crash-recovery
    • Omission Faults
    • Arbitrary Failures Mode (Byzantine)
    

    View Slide

41. LIVENESS AND
    SAFETY
    

    View Slide

42. LIVENESS AND SAFETY
    PROPERTIES OF ALGORITHMS
    

    View Slide

43. SAFETY
    Some “bad” thing does not
    happens during execution
    

    View Slide

44. SAFETY
    “Communication links should not
    invent messages out of thin air”
    

    View Slide

45. LIVENESS
    A “good” thing happens during
    execution
    

    View Slide

46. LIVENESS
    “A destination process eventually
    delivers the message”
    

    View Slide

47. LET’S TAKE A LOOK
    AT FLP1
    1 - Fischer, Lynch, Paterson
    

    View Slide

48. View Slide

49. IMPOSSIBILITY OF DISTRIBUTED
    CONSENSUS WITH ONE FAULTY
    PROCESS
    

    View Slide

50. IMPOSSIBILITY OF DISTRIBUTED
    CONSENSUS WITH ONE FAULTY
    PROCESS
    

    View Slide

51. IMPOSSIBILITY OF DISTRIBUTED
    CONSENSUS WITH ONE FAULTY
    PROCESS
    

    View Slide

52. IMPOSSIBILITY OF DISTRIBUTED
    CONSENSUS WITH ONE FAULTY
    PROCESS
    

    View Slide

53. IMPOSSIBILITY OF DISTRIBUTED
    CONSENSUS WITH ONE FAULTY
    PROCESS
    

    View Slide

54. WHAT’S CONSENSUS
    ANYWAY?
    

    View Slide

55. “THE CONSENSUS
    PROBLEM IS A PARADIGM
    OF AGREEMENT
    PROBLEMS”
    https://dl.acm.org/citation.cfm?id=1052796.1052806
    

    View Slide

56. PROPERTIES OF
    CONSENSUS
    

    View Slide

57. PROPERTIES OF
    CONSENSUS
    • C-Termination: Every correct process eventually decides on some value
    

    View Slide

58. PROPERTIES OF
    CONSENSUS
    • C-Termination: Every correct process eventually decides on some value
    • C-Validity: If a process decides v, then v was proposed by some process
    

    View Slide

59. PROPERTIES OF
    CONSENSUS
    • C-Termination: Every correct process eventually decides on some value
    • C-Validity: If a process decides v, then v was proposed by some process
    • C-Agreement: No two correct processes decide differently
    

    View Slide

60. PROPERTIES OF UNIFORM
    CONSENSUS
    • C-Termination: Every correct process eventually decides on some value
    • C-Validity: If a process decides v, then v was proposed by some process
    • C-Agreement: No two correct processes decide differently
    • C-Uniform Agreement: No two processes (correct or not) decide
    differently.
    

    View Slide

61. WE NEED CONSENSUS
    WHEN:
    A SET OF PROCESSES
    HAVE TO AGREE TO TAKE
    A COMMON ACTION
    

    View Slide

62. WE NEED CONSENSUS
    WHEN:
    A SET OF PROCESSES
    HAVE TO AGREE TO TAKE
    A COMMON ACTION
    Atomic
    Broadcast
    

    View Slide

63. WE NEED CONSENSUS
    WHEN:
    A SET OF PROCESSES
    HAVE TO AGREE TO TAKE
    A COMMON ACTION
    Atomic
    Broadcast
    Group
    Membership
    

    View Slide

64. ATOMIC BROADCAST
    “CORRECT PROCESSES
    DELIVER THE SAME SET OF
    MESSAGES IN THE SAME
    ORDER”
    

    View Slide

65. FLP TELLS US THAT IF
    CONSENSUS CANNOT BE
    ACHIEVED, THEN ATOMIC
    BROADCAST OR GROUP
    MEMBERSHIP CANNOT BE
    ACHIEVED EITHER
    

    View Slide

66. SO, WE PACK OUR BAGS
    AND GO?
    NOTHING TO SEE HERE?
    

    View Slide

67. STUMBLING OVER
    CONSENSUS RESEARCH:
    MISUNDERSTANDING AND
    ISSUES
    Marcos K. Aguilera
    

    View Slide

68. FAILURE
    DETECTORS
    

    View Slide

69. View Slide

70. FAILURE DETECTORS
    

    View Slide

71. FAILURE DETECTORS
    • External process
    

    View Slide

72. FAILURE DETECTORS
    • External process
    • Provides information about suspected processes
    

    View Slide

73. FAILURE DETECTORS
    • External process
    • Provides information about suspected processes
    • Completeness property (crashed processes are
    detected)
    

    View Slide

74. FAILURE DETECTORS
    • External process
    • Provides information about suspected processes
    • Completeness property (crashed processes are
    detected)
    • Accuracy (correct process are never suspected)
    

    View Slide

75. “RUB SOME PERFECT
    FAILURE DETECTOR
    ON IT”
    

    View Slide

76. http://www.amazon.com/Introduction-Reliable-Secure-
    Distributed-Programming/dp/3642152597
    PERFECT FAILURE
    DETECTOR
    

    View Slide

77. EVENTUALLY ACCURATE
    FAILURE DETECTOR
    

    View Slide

78. EVENTUALLY ACCURATE
    FAILURE DETECTOR
    • Strong Completeness: Eventually, every
    process that crashes is permanently
    suspected by every correct process.
    

    View Slide

79. EVENTUALLY ACCURATE
    FAILURE DETECTOR
    • Strong Completeness: Eventually, every
    process that crashes is permanently
    suspected by every correct process.
    • Eventual Weak Accuracy: There is a time
    after which some correct process is never
    suspected by the correct processes.
    

    View Slide

80. EVENTUALLY ACCURATE
    FAILURE DETECTOR
    • Strong Completeness: Eventually, every
    process that crashes is permanently
    suspected by every correct process.
    • Eventual Weak Accuracy: There is a time
    after which some correct process is never
    suspected by the correct processes.
    http://dl.acm.org/citation.cfm?id=1052806
    

    View Slide

81. View Slide

82. QUORUMS
    

    View Slide

83. TL;DR:
    INTERSECTING
    SETS
    

    View Slide

84. “A QUORUM IN A SYSTEM WITH N
    CRASH-FAULT PROCESS ABSTRACTIONS
    […] IS ANY MAJORITY OF
    PROCESSES, I.E., ANY SET OF MORE
    THAN N/2 PROCESSES”
    QUORUMS
    

    View Slide

85. “IF F < N/2 PROCESSES FAIL BY
    CRASHING, THERE IS ALWAYS AT
    LEAST ONE QUORUM OF
    NONCRASHED PROCESSES IN SUCH
    SYSTEMS”
    QUORUMS
    

    View Slide

86. CONSISTENCY
    

    View Slide

87. View Slide

88. CONCURRENT
    FIFO QUEUE
    

    View Slide

89. CONSISTENCY
    CONDITIONS
    

    View Slide

90. CONSISTENCY
    CONDITIONS
    • Atomic Consistency (Linearizabilty)
    

    View Slide

91. CONSISTENCY
    CONDITIONS
    • Atomic Consistency (Linearizabilty)
    • Sequential Consistency
    

    View Slide

92. CONSISTENCY
    CONDITIONS
    • Atomic Consistency (Linearizabilty)
    • Sequential Consistency
    • Causal Consistency
    

    View Slide

93. CONSISTENCY
    CONDITIONS
    • Atomic Consistency (Linearizabilty)
    • Sequential Consistency
    • Causal Consistency
    https://aphyr.com/posts/313-strong-consistency-
    models
    

    View Slide

94. LINEARIZABILTY
    http://www.amazon.com/Distributed-Algorithms-Message-Passing-Systems-Michel/dp/
    3642381227/
    

    View Slide

95. LINEARIZABILTY
    http://www.amazon.com/Distributed-Algorithms-Message-Passing-Systems-Michel/dp/
    3642381227/
    

    View Slide

96. SOME BOOKS
    

    View Slide

97. http://www.amazon.com/Distributed-Algorithms-
    Message-Passing-Systems-Michel/dp/3642381227/
    

    View Slide

98. http://www.amazon.com/Fault-tolerant-Agreement-
    Synchronous-Message-passing-Distributed/dp/
    1608455254/
    

    View Slide

99. http://www.amazon.com/Communication-Abstractions-
    Fault-tolerant-Asynchronous-Distributed/dp/160845293X/
    

    View Slide

100. http://www.amazon.com/Distributed-Algorithms-
     Kaufmann-Management-Systems/dp/1558603484/
     

     View Slide

101. http://www.amazon.com/Introduction-Reliable-Secure-
     Distributed-Programming/dp/3642152597
     

     View Slide

102. http://www.amazon.com/Guide-Reliable-Distributed-
     Systems-High-Assurance/dp/1447124154/
     

     View Slide

103. http://www.amazon.com/Replication-Practice-Lecture-
     Computer-Theoretical/dp/3642112935/
     

     View Slide

104. FINDING NON
     PAYWALLED PAPERS
     

     View Slide

105. CONCLUSION
     

     View Slide

106. CONCLUSION
     • Deep Rabbit Hole
     

     View Slide

107. CONCLUSION
     • Deep Rabbit Hole
     • Computing Science where Science is Still a Thing™
     

     View Slide

108. CONCLUSION
     • Deep Rabbit Hole
     • Computing Science where Science is Still a Thing™
     • History of the Field Matters
     

     View Slide

109. CONCLUSION
     • Deep Rabbit Hole
     • Computing Science where Science is Still a Thing™
     • History of the Field Matters
     • Read, read, read
     

     View Slide

110. THANKS!
     @old_sound
     

     View Slide