Operations without the Operating System

Talk from the Operability conference in London about fundamental problems with the evolution of infrastructure, and how new operating systems and unikernels might provide a solutions.

Some truth and lots of speculation about the future of operations.

Transcript

1. Operations without the
   Operating System
   Puppet Labs
   Gareth Rushgrove
   Ops People and Future Technology
   

   View Slide

2. Gareth Rushgrove
   @garethr
   

   View Slide

3. Gareth Rushgrove
   

   View Slide

4. First, some quotes
   

   View Slide

5. These systems had never
   been designed with security in
   mind and now were becoming
   hugely complex
   Gareth Rushgrove
   

   View Slide

6. The dense complexity and
   increasing length of computer
   code as computers where asked
   to do more and more would make
   malicious code increasingly hard
   to find
   Gareth Rushgrove
   

   View Slide

7. Written in 2014
   Gareth Rushgrove
   

   View Slide

8. Not about Heartbleed
   Gareth Rushgrove
   

   View Slide

9. In a history book about the 1970s
   Gareth Rushgrove
   Intercept: The Secret History of Computers and Spies
   by Gordon Corera. Specifically about the Anderson Report, 1972
   

   View Slide

10. This talk
    

    View Slide

11. Lots of new operating systems
    have appeared in the last year
    Gareth Rushgrove
    

    View Slide

12. Speculation Klaxon
    

    View Slide

13. The age of the general purpose
    operating system is over
    Gareth Rushgrove
    

    View Slide

14. What does that mean for
    operators?
    Gareth Rushgrove
    

    View Slide

15. A New Breed of OS
    Smaller, faster, more productive?
    

    View Slide

16. What server operating systems
    are you running?
    Gareth Rushgrove
    

    View Slide

17. Gareth Rushgrove
    I’m guessing most people said:
    

    View Slide

18. Gareth Rushgrove
    Someone will try and be smart and say:
    

    View Slide

19. Gareth Rushgrove
    And someone will hide the fact they’re running:
    

    View Slide

20. But elsewhere on the internet…
    Gareth Rushgrove
    

    View Slide

21. Gareth Rushgrove
    From RedHat, uses
    RPM-OSTree under
    the hood
    

    View Slide

22. Gareth Rushgrove
    Built-in service-discovery
    and clustering support
    

    View Slide

23. Gareth Rushgrove
    Snappy from Ubuntu,
    basically replaces apt-get
    with containers
    

    View Slide

24. Gareth Rushgrove
    Docker running on
    Docker running as PID 1
    

    View Slide

25. Gareth Rushgrove
    Nano is a tiny alternative
    to Windows Server
    6
    

    View Slide

26. Gareth Rushgrove
    Bonneville provides
    containers that are really
    virtual machines
    

    View Slide

27. Gareth Rushgrove
    Clear Linux provides
    hardware isolation
    guarantees to containers
    

    View Slide

28. Common themes
    Gareth Rushgrove
    

    View Slide

29. Cluster native
    Read-only file systems
    Transactional updates
    Integrated with containers
    Gareth Rushgrove
    

    View Slide

30. We’ve seen some of this
    story before
    Gareth Rushgrove
    

    View Slide

31. Gareth Rushgrove
    

    View Slide

32. ESXi, XenServer
    Gareth Rushgrove
    

    View Slide

33. Gareth Rushgrove
    

    View Slide

34. What plays out the same
    and what different is
    interesting
    Gareth Rushgrove
    Note, everyone wants to be the VMware of containers
    

    View Slide

35. Why The Interest in New
    Operating Systems
    And why now?
    

    View Slide

36. Lots of homogeneous workloads
    Gareth Rushgrove
    

    View Slide

37. Security is front page news
    Gareth Rushgrove
    

    View Slide

38. I just want somewhere to run my
    containers
    Gareth Rushgrove
    

    View Slide

39. Size as a proxy for complexity
    Gareth Rushgrove
    Can you name the 184 default packages in Ubuntu?
    

    View Slide

40. Utilisation matters
    Gareth Rushgrove
    Only to large organisations who pay for software mind
    

    View Slide

41. Increasingly interacting with
    higher level abstractions anyway
    Gareth Rushgrove
    Mesos, Kubernetes, Cloud Foundry
    

    View Slide

42. Piling up Abstractions
    A brief history of infrastructure
    

    View Slide

43. Gareth Rushgrove
    Operating system
    Hardware
    Application
    Once upon a time…
    

    View Slide

44. Gareth Rushgrove
    Operating system
    Hardware
    Runtime
    Application
    But your application might
    need a runtime so lets
    add that
    

    View Slide

45. Gareth Rushgrove
    Operating system
    Hardware
    Runtime
    Application
    Application Application
    Application
    Application Application
    Runtime
    But hardware is
    expensive so lets run
    multiple applications
    

    View Slide

46. Gareth Rushgrove
    Operating system
    Hypervisor
    Hardware
    Runtime
    Application
    Applications need
    isolating so lets use
    virtualisation
    

    View Slide

47. Gareth Rushgrove
    Operating system
    Hypervisor
    Hardware
    Runtime
    Application
    Operating system
    Runtime
    Application
    Operating system
    Runtime
    Application
    Run multiple virtual
    machines! Each with
    there own copy of the OS
    

    View Slide

48. Gareth Rushgrove
    Operating system
    Hypervisor
    Hardware
    Runtime
    Application
    Operating system
    Runtime
    Application Application
    Application Application
    But the overhead of
    virtualisation is expensive
    so run multiple apps per VM
    

    View Slide

49. Gareth Rushgrove
    Operating system
    Hypervisor
    Hardware
    Container
    Operating system
    Runtime
    Application
    Container runtime
    I heard you like
    containers
    

    View Slide

50. Gareth Rushgrove
    Operating system
    Hypervisor
    Hardware
    Container
    A different OS
    Runtime
    Application
    Container runtime
    Container
    Operating system
    Runtime
    Application
    Container
    Even more OS
    Runtime
    Application
    

    View Slide

51. Hypervisor
    Hardware
    Gareth Rushgrove
    Operating system
    Container
    Static binary
    Container runtime
    Container
    Static binary
    Container
    Static binary
    Cool folks use static
    binaries and scratch
    containers
    

    View Slide

52. Gareth Rushgrove
    Operating system
    Hardware
    Container
    Static binary
    Container runtime
    Container
    Static binary
    Container
    Static binary
    Don’t need
    virtualisation isolation
    guarantees?
    

    View Slide

53. Unikernels
    A library operating system
    

    View Slide

54. What if there is no OS
    above the hypervisor?
    Gareth Rushgrove
    

    View Slide

55. Gareth Rushgrove
    

    View Slide

56. Gareth Rushgrove
    Unikernel
    Hypervisor
    Hardware
    Unikernels compile your
    application to a kernel, which
    can run on a hypervisor
    

    View Slide

57. Compile your application
    into a Kernel
    Gareth Rushgrove
    No userspace
    

    View Slide

58. Only include the capabilities/
    libraries you need
    Gareth Rushgrove
    No bash/ssh/sh/apt unless you explicitly include it
    

    View Slide

59. Hypervisor/hardware isolation
    Smaller attack surface area
    Running less code
    Enforced immutability
    No default remote access
    Gareth Rushgrove
    

    View Slide

60. The promise of containers, but
    without needing to pretend the
    intermediary OS doesn’t exist
    Gareth Rushgrove
    

    View Slide

61. Gareth Rushgrove
    MirageOS
    

    View Slide

62. Gareth Rushgrove
    HaLVM
    

    View Slide

63. Gareth Rushgrove
    Rump Kernel
    

    View Slide

64. Gareth Rushgrove
    LING
    

    View Slide

65. Gareth Rushgrove
    

    View Slide

66. Gareth Rushgrove
    

    View Slide

67. What Happens to
    Operators?
    Technical operators anyway, service
    
    management folks are safe for now
    

    View Slide

68. Hypervisor as the platform
    Gareth Rushgrove
    1
    

    View Slide

69. Separate team running a
    hypervisor as a service
    Gareth Rushgrove
    

    View Slide

70. Separate team company
    running a hypervisor as a
    service
    Gareth Rushgrove
    

    View Slide

71. How many AWS users are
    Xen experts?
    Gareth Rushgrove
    

    View Slide

72. It happened with Type 1
    hypervisors, I posit it will
    happen for Type 2 as well
    Gareth Rushgrove
    

    View Slide

73. Everything else as an application
    Gareth Rushgrove
    2
    

    View Slide

74. Firewalls as an application
    Gareth Rushgrove
    

    View Slide

75. Proxies as an application
    Gareth Rushgrove
    

    View Slide

76. Network switches as an application
    Gareth Rushgrove
    

    View Slide

77. Intrusion detection as an
    application
    Gareth Rushgrove
    

    View Slide

78. Remote shell as an application
    Gareth Rushgrove
    

    View Slide

79. Everyone not running
    the hypervisor is an
    application developer
    Gareth Rushgrove
    

    View Slide

80. Standards, standards and
    defacto standards
    Gareth Rushgrove
    3
    

    View Slide

81. We need to start agreeing
    Gareth Rushgrove
    I don’t have time here for a rant about the contributors dilemma,
    initiation bias and how prototypes never die on GitHub
    

    View Slide

82. Platforms
    Gareth Rushgrove
    

    View Slide

83. Containers
    Gareth Rushgrove
    

    View Slide

84. Monitoring
    Gareth Rushgrove
    ?
    

    View Slide

85. Gareth Rushgrove
    StatsD
    

    View Slide

86. Gareth Rushgrove
    Metrics 2.0
    

    View Slide

87. Publish more schemas and fewer
    incompatible duplicate
    implementations
    Gareth Rushgrove
    We can get to actual Standards after some agreement
    

    View Slide

88. Infrastructure is code
    Gareth Rushgrove
    4
    

    View Slide

89. Infrastructure as algorithms
    Gareth Rushgrove
    

    View Slide

90. Schedulers
    Gareth Rushgrove
    

    View Slide

91. Back pressure
    Gareth Rushgrove
    

    View Slide

92. Autoscaling
    Gareth Rushgrove
    

    View Slide

93. Composition/configuration
    Gareth Rushgrove
    

    View Slide

94. Configuration with static data
    won’t be enough
    Gareth Rushgrove
    

    View Slide

95. Learn to programme. This time
    we mean it
    Gareth Rushgrove
    If you want this future we need to build it. Probably with
    static types and safer programming languages
    

    View Slide

96. Revolution not evolution
    Gareth Rushgrove
    5
    

    View Slide

97. The distance between old
    infrastructure and new will
    be huge
    Gareth Rushgrove
    

    View Slide

98. Moving between the paradigms
    will be bigger than moving
    between Linux and Windows
    today
    Gareth Rushgrove
    

    View Slide

99. Not about tools. About models of
    interaction and the skills required
    to operate
    Gareth Rushgrove
    

    View Slide

100. We either fix our tech, or the
     security challenge will result in
     laws and policy that try and fix
     it for us
     Gareth Rushgrove
     I remember when it was legal to run software I wrote myself
     without cyber insurance. Ah, happy days.
     

     View Slide

101. Gareth Rushgrove
     Remember. Blade
     Runner is a movie about
     the Government running
     around fixing bugs in
     software
     

     View Slide

102. Conclusions
     If all you remember from this talk is…
     

     View Slide

103. We still have fundamental
     problems that date back 40 years
     Gareth Rushgrove
     

     View Slide

104. We can build better infrastructure
     Gareth Rushgrove
     

     View Slide

105. Doing so might mean going down
     different evolutionary paths
     Gareth Rushgrove
     

     View Slide

106. We know many of the patterns
     we want, but we’re mainly
     working backwards
     Gareth Rushgrove
     Stripping down our current operating systems isn’t sufficient
     

     View Slide

107. This means throwing away things
     we care about
     Gareth Rushgrove
     What would you say if I said we don’t need Linux for the general case
     

     View Slide

108. Collaborate on hard problems,
     rather than marvelling about how
     easy it is to build your own
     Docker PaaS
     Gareth Rushgrove
     

     View Slide

109. Questions?
     And thanks for listening
     

     View Slide