Upgrade to Pro — share decks privately, control downloads, hide ads and more …

The Ecological Impact of Compute

Seth Vargo
August 19, 2016

The Ecological Impact of Compute

As compute becomes cheaper and more readily available, the costs to our environment cannot go unnoticed. We are increasing energy use at exponential rates by over-provisioning requirements, under-utliizing resources, and clicking buttons in UIs. As a business owner, you want to save costs, but what are the environmental impacts services like AWS and GCE are having on our planet? Tools like Nomad, Kubernetes, and Mesos allow organizations to maximize resource utilization, but can schedulers also be used to help save our planet?

Seth Vargo

August 19, 2016
Tweet

More Decks by Seth Vargo

Other Decks in Technology

Transcript

  1. https://upload.wikimedia.org/wikipedia/commons/d/d7/Wikimedia_Foundation_Servers-8055_24.jpg
    Ecological
    Compute
    IMPACT OF
    THE

    View full-size slide

  2. @SETHVARGO
      

    View full-size slide

  3. https://upload.wikimedia.org/wikipedia/commons/5/5b/Italian_legal_hardback_books_open.jpg
    DISCLAIMERS
    BUT FIRST A FEW

    View full-size slide

  4. https://upload.wikimedia.org/wikipedia/commons/4/4a/March_Against_Monsanto_Vancouver.jpg
    I AM NOT AN ENVIRONMENTAL ACTIVIST

    View full-size slide

  5. https://upload.wikimedia.org/wikipedia/commons/1/1f/Lincoln_mkc_front.jpg
    19 MPG

    View full-size slide

  6. OOPS
    https://upload.wikimedia.org/wikipedia/commons/8/87/Nest_Learning_Thermostat_(cropped).JPG

    View full-size slide

  7. https://upload.wikimedia.org/wikipedia/commons/5/51/Old_book_-_Basking_Ridge_Historical_Society.jpg
    BACKGROUND
    AND NOW SOME

    View full-size slide

  8. https://upload.wikimedia.org/wikipedia/commons/c/c3/Internet_map_4096.png

    View full-size slide

  9. EVERY 60 SECONDS
    15,000 MUSIC DOWNLOADS

    350,000 TWITTER TWEETS

    5 MILLION GOOGLE SEARCHES

    204 MILLION EMAIL MESSAGES

    1.8 MILLION FACEBOOK LIKES

    $272,000 AMAZON PURCHASES

    View full-size slide

  10. EVERY 60 SECONDS
    204 MILLION
    5 MILLION
    1.8 MILLION
    350,000
    $272,000
    15,000
    EMAIL MESSAGES
    GOOGLE SEARCHES
    FACEBOOK LIKES
    TWITTER TWEETS
    AMAZON PURCHASES
    MUSIC DOWNLOADS






    View full-size slide

  11. 91 BILLION KILOWATT-HOURS
    https://static.pexels.com/photos/7000/fre-sonneveld-powerlines.jpg

    View full-size slide

  12. https://upload.wikimedia.org/wikipedia/commons/d/d6/Coal_power_plant_Datteln_2_Crop1.png
    500-MEGAWATT POWER PLANTS

    View full-size slide

  13. https://upload.wikimedia.org/wikipedia/commons/d/d6/Coal_power_plant_Datteln_2_Crop1.png
    500-MEGAWATT POWER PLANTS
    34 TIMES

    View full-size slide

  14. Russia
    Japan
    India
    Germ
    any
    Canada
    France
    Brazil
    S. Korea UK
    Italy
    Spain
    Taiwan
    S. Africa
    Australia
    M
    exico
    Suppresses China and USA

    View full-size slide

  15. Russia
    Japan
    India
    Germ
    any
    Canada
    France
    Brazil
    S. Korea UK
    Italy
    Spain
    Taiwan
    S. Africa
    Australia
    M
    exico
    Suppresses China and USA

    View full-size slide

  16. 100 MILLION TONS OF CARBON POLLUTION
    https://upload.wikimedia.org/wikipedia/commons/c/cb/Automobile_exhaust_gas.jpg

    View full-size slide

  17. 100 MILLION TONS OF CARBON POLLUTION
    https://upload.wikimedia.org/wikipedia/commons/c/cb/Automobile_exhaust_gas.jpg
    26.5 MILLION AUTOMOBILES

    View full-size slide

  18. 100 MILLION TONS OF CARBON POLLUTION
    https://upload.wikimedia.org/wikipedia/commons/c/cb/Automobile_exhaust_gas.jpg
    26.5 MILLION AUTOMOBILES
    470 MILLION TREES

    View full-size slide

  19. https://i.vimeocdn.com/video/526825393_1280x960.jpg
    SOLUTIONS
    NOT SO OBVIOUS

    View full-size slide

  20. INCREASE IN
    PUBLIC CLOUD USE
    INCREASE IN
    CARBON FOOTPRINT

    View full-size slide

  21. CORRELATION
    CAUSATION
    INCREASE IN
    PUBLIC CLOUD USE
    INCREASE IN
    CARBON FOOTPRINT

    View full-size slide

  22. http://www.thegraphicrecorder.com/wp-content/uploads/2012/01/CorrelationCausationFinal1.jpg

    View full-size slide

  23. 4 TYPES OF DATA CENTERS
    https://upload.wikimedia.org/wikipedia/commons/6/67/Inside_Suite.jpg

    View full-size slide

  24. WHOLESALE CO-LO
    CUSTOMER

    View full-size slide

  25. WHOLESALE CO-LO RETAIL CO-LO
    CUSTOMER CUSTOMER 2
    CUST 1 C3

    View full-size slide

  26. WHOLESALE CO-LO RETAIL CO-LO
    MANAGED HOSTING
    CUSTOMER CUSTOMER 2
    CUST 1 C3
    CUSTOMER 2
    CUST 1 C3

    View full-size slide

  27. WHOLESALE CO-LO RETAIL CO-LO
    MANAGED HOSTING CLOUD PROVIDER
    CUSTOMER CUSTOMER 2
    CUST 1 C3
    CUSTOMER 2
    CUST 1 C3 CUSTOMERS

    View full-size slide

  28. WHOLESALE CO-LO RETAIL CO-LO
    MANAGED HOSTING CLOUD PROVIDER
    49% 27%
    19% 5%

    View full-size slide

  29. Cloud Provider
    5%
    Managed Hosting
    19%
    Retail Co-lo
    27%
    Wholesale Co-lo
    49%

    View full-size slide

  30. CAUSES OF INEFFICIENCY
    http://www.trbimg.com/img-57322fc1/turbine/ct-tsa-airport-security-lines-20160510

    View full-size slide

  31. OVER-PROVISIONING OF IT RESOURCES
    https://static.pexels.com/photos/7000/fre-sonneveld-powerlines.jpg
    #1

    View full-size slide

  32. PEAK PROVISIONING
    https://upload.wikimedia.org/wikipedia/commons/8/86/Peak_of_the_Matterhorn,_seen_from_Zermatt,_Switzerland.jpg

    View full-size slide

  33. 1 2 3 4 5
    8 9 10 11 12
    7
    6
    15 16 17 18 19
    14
    13
    22 23 24 25 26
    21
    20
    29 30
    28
    27
    BLACK FRIDAY
    NEW SERVERS
    These numbers are fictitious and are not based on statistical evidence; shown for demonstrative purposes only

    View full-size slide

  34. 1 2 3 4 5
    8 9 10 11 12
    7
    6
    15 16 17 18 19
    14
    13
    22 23 24 25 26
    21
    20
    29 30
    28
    27
    BLACK FRIDAY
    NEW SERVERS
    
    These numbers are fictitious and are not based on statistical evidence; shown for demonstrative purposes only

    View full-size slide

  35. 1 2 3 4 5
    8 9 10 11 12
    7
    6
    15 16 17 18 19
    14
    13
    22 23 24 25 26
    21
    20
    29 30
    28
    27
    BLACK FRIDAY
    NEW SERVERS
     – $500 – $500
    – $500 – $500 – $500 – $500 – $500 – $500 – $500
    – $500 – $500 – $500 – $500 – $500
    These numbers are fictitious and are not based on statistical evidence; shown for demonstrative purposes only

    View full-size slide

  36. 1 2 3 4 5
    8 9 10 11 12
    7
    6
    15 16 17 18 19
    14
    13
    22 23 24 25 26
    21
    20
    29 30
    28
    27
    BLACK FRIDAY
    NEW SERVERS
     – $500 – $500
    – $500 – $500 – $500 – $500 – $500 – $500 – $500
    – $500 – $500 – $500 – $500 – $500 + $10,000
    These numbers are fictitious and are not based on statistical evidence; shown for demonstrative purposes only

    View full-size slide

  37. 1 2 3 4 5
    8 9 10 11 12
    7
    6
    15 16 17 18 19
    14
    13
    22 23 24 25 26
    21
    20
    29 30
    28
    27
    BLACK FRIDAY
    NEW SERVERS
     – $500 – $500
    – $500 – $500 – $500 – $500 – $500 – $500 – $500
    – $500 – $500 – $500 – $500 – $500 + $10,000 – $500
    – $500 – $500 – $500 – $500
    These numbers are fictitious and are not based on statistical evidence; shown for demonstrative purposes only

    View full-size slide

  38. INVALID IT PROCUREMENT FOCUS
    #2

    View full-size slide

  39. HOW IT
    PROCUREMENT DECISIONS
    SHOULD BE ANALYZED
    http://www.policonomics.com/wp-content/uploads/Cost-analysis-long-term.jpg

    View full-size slide

  40. HOW IT
    PROCUREMENT DECISIONS
    ARE ACTUALLY MADE
    http://acclinet.com/images/stories/sun-dell-ad.jpg

    View full-size slide

  41. HOW IT
    PROCUREMENT DECISIONS
    ARE ACTUALLY MADE
    https://i.ytimg.com/vi/J2WNUXTRNes/maxresdefault.jpg

    View full-size slide

  42. HOW IT
    PROCUREMENT DECISIONS
    ARE ACTUALLY MADE
    http://ad-design.966v.com/static_images/20160724/cd767d958d873b6ab96082b93951e53865607b3ea4ca315e0aeff539.jpg

    View full-size slide

  43. UNUSED "ZOMBIE" SERVERS
    https://upload.wikimedia.org/wikipedia/commons/1/17/2007-04-07_-_London_-_Flashmob_-_Fleshmob_-_Zombie_Walk_-_Zombies_(4889850244).jpg
    #3

    View full-size slide

  44. An estimated 20-30% of servers in
    [wholesale] data centers today are idle,
    obsolete, or unused but are still plugged
    in and consuming energy doing
    nothing.

    View full-size slide

  45. https://www.youtube.com/watch?v=95y0tPseZRE

    View full-size slide

  46. IT managers cannot identify owners for
    15-30% of the servers, but are reluctant
    to decommission equipment, fearing
    potential impacts on business or
    application functions.

    View full-size slide

  47. https://journal.uptimeinstitute.com/wp-content/uploads/2014/08/server-roundup-cover-image.jpg

    View full-size slide

  48. https://journal.uptimeinstitute.com/wp-content/uploads/2014/08/server-roundup-cover-image.jpg
    DECOMMISSIONED MORE THAN 9,500 SERVERS
    $5M IN ANNUAL ENERGY SAVINGS
    5 MEGAWATTS IN IT LOAD
    4 MEGAWATTS OF ASSOCIATED COOLING

    View full-size slide

  49. https://journal.uptimeinstitute.com/wp-content/uploads/2014/08/server-roundup-cover-image.jpg
    MISALIGNMENT OF RESPONSIBILITIES
    ELECTRIC COSTS PAID BY A DIFFERENT DEPARTMENT
    NO INCENTIVE FOR DATA CENTER TEAMS TO REDUCE LOAD

    View full-size slide

  50. https://upload.wikimedia.org/wikipedia/commons/d/df/Port_Authority_bus_Pittsburgh_3216.jpg
    LACK OF STANDARD UTILIZATION METRIC
    #4

    View full-size slide

  51. 80% CPU 60% MEM
     

    View full-size slide

  52. 80% CPU 60% MEM 1500 IOPS
     

    View full-size slide

  53. 80% CPU 60% MEM 1500 IOPS ?
     

    × ÷ ≈

    View full-size slide

  54. http://i.huffpost.com/gen/1821469/images/o-LANGUAGES-facebook.jpg

    View full-size slide

  55. https://upload.wikimedia.org/wikipedia/commons/4/43/Okinawan_High_Schools_compete_in_race_at_MCAS_Futenma_141001-M-PU373-021.jpg
    COMPETING PRIORITIES FOR EFFICIENCY
    #5

    View full-size slide

  56. Efficiency
    4%
    Deployments
    24%
    Security
    24%
    Reliability
    24%
    Availability
    24%

    View full-size slide

  57. Deployments
    25%
    Security
    25%
    Reliability
    25%
    Availability
    25%

    View full-size slide

  58. http://gssolutions.com/wp-content/uploads/2014/12/cloudit.jpg
    CLOUD
    TO THE

    View full-size slide

  59. WE DON'T GET THE SAME PERFORMANCE ON THE CLOUD
    MANY ORGANIZATIONS DO NOT UNDERSTAND THEIR APPLICATION
    REQUIREMENTS
    MYTH
    FACT

    View full-size slide

  60. THE CLOUD IS INSECURE – WE HAVE TO BE HIPPA/PCI COMPLIANT
    CLOUD PROVIDERS PAY MILLIONS OF DOLLARS AND HIRE THE MOST
    EXPERIENCED PROFESSIONALS ON THESE TOPICS.
    ARE YOU BETTER IF COMPLIANCE IS SOMEONE'S PART-TIME JOB?
    MYTH
    FACT

    View full-size slide

  61. THE CLOUD IS TOO EXPENSIVE
    FOR MOST USE CASES, IT'S PROBABLY CHEAPER AND MORE FLEXIBLE
    MYTH
    FACT

    View full-size slide

  62. WE RUN ON OUR OWN BARE METAL AND ACHIEVE 80% UTILIZATION USING
    A SCHEDULER AND ROUTINELY AUDIT SERVER USAGE USING AN INVENTORY
    MANAGEMENT SYSTEM.
    THIS TALK IS NOT FOR YOU, BUT
    MYTH
    FACT

    View full-size slide

  63. I AM ENERGY AND RESOURCE EFFICIENT BECAUSE I AM USING A
    CLOUD PROVIDER
    YOU ARE IMPOSING LESS OF AN ENVIRONMENTAL IMPACT, BUT YOU
    MAY NOT BE USING YOUR POOL OF RESOURCES AT BEST CAPACITY
    MYTH
    FACT

    View full-size slide

  64. http://i.huffpost.com/gen/1821469/images/o-LANGUAGES-facebook.jpg
    LIVE DEMO

    View full-size slide

  65. https://upload.wikimedia.org/wikipedia/commons/8/82/Cloture_grande_faune_A73A.jpg
    UTILIZATION
    MAXIMIZING RESOURCE

    View full-size slide

  66. HUMANS MAXIMIZE TIME WITH THESE
    https://upload.wikimedia.org/wikipedia/commons/0/01/February_calendar.jpg

    View full-size slide

  67. sched·ul·er (human)
    https://upload.wikimedia.org/wikipedia/commons/0/01/February_calendar.jpg
    a person or machine that organizes or
    maintains schedules

    View full-size slide

  68. sched·ul·er (computer)
    https://upload.wikimedia.org/wikipedia/commons/0/01/February_calendar.jpg
    a program that arranges computer
    operations into an appropriate sequence

    View full-size slide

  69. https://cdn.shopify.com/s/files/1/0167/3936/files/04-MS-Excel-4-0-Office-4-0_large.gif

    View full-size slide



  70. Operator
    Datacenter

    View full-size slide



  71. Operator
    Datacenter
    Skywalker Vader Leia Solo

    View full-size slide



  72. Operator
    Datacenter
    PYTHON
    PYTHON
    GOLANG
    GOLANG
    GOLANG
    Skywalker Vader Leia Solo

    View full-size slide



  73. Operator
    Datacenter
    RUBY PYTHON
    PYTHON
    PYTHON
    GOLANG
    GOLANG
    GOLANG
    GOLANG
    NODE
    Skywalker Vader Leia Solo

    View full-size slide



  74. Operator
    Datacenter
    RUBY PYTHON
    PYTHON
    PYTHON
    GOLANG
    GOLANG
    GOLANG
    GOLANG
    NODE
    Skywalker Vader Leia Solo
    RUBY
    VADER
    LEIA
    SOLO
    192.168.1.4
    192.168.1.5
    192.168.1.7
    192.168.1.253
    88:45:13:B6:87:C4
    94:CE:4F:C8:54:C3
    CA:9A:3D:7F:8B:CB
    72:30:9C:0D:1E:74
    Randomly kills applications

    View full-size slide



  75. Operator
    Datacenter
    RUBY PYTHON
    PYTHON
    PYTHON
    GOLANG
    GOLANG
    GOLANG
    GOLANG
    NODE
    Skywalker Leia Solo
    RUBY
    VADER
    LEIA
    SOLO
    192.168.1.4
    192.168.1.5
    192.168.1.7
    192.168.1.253
    88:45:13:B6:87:C4
    94:CE:4F:C8:54:C3
    CA:9A:3D:7F:8B:CB
    72:30:9C:0D:1E:74
    Randomly kills applications
    F
    F
    Vader

    View full-size slide



  76. Operator
    Datacenter
    RUBY PYTHON
    PYTHON
    PYTHON
    GOLANG
    GOLANG
    GOLANG
    GOLANG
    NODE
    Skywalker Leia Solo
    RUBY
    VADER
    LEIA
    SOLO
    192.168.1.4
    192.168.1.5
    192.168.1.7
    192.168.1.253
    88:45:13:B6:87:C4
    94:CE:4F:C8:54:C3
    CA:9A:3D:7F:8B:CB
    72:30:9C:0D:1E:74
    Randomly kills applications
    F
    F
    Vader
    PYTHON
    PYTHON
    PYTHON

    View full-size slide



  77. Operator
    Datacenter
    RUBY GOLANG
    GOLANG
    GOLANG
    GOLANG
    NODE
    Skywalker Leia Solo
    RUBY
    VADER
    LEIA
    SOLO
    192.168.1.4
    192.168.1.5
    192.168.1.7
    192.168.1.253
    88:45:13:B6:87:C4
    94:CE:4F:C8:54:C3
    CA:9A:3D:7F:8B:CB
    72:30:9C:0D:1E:74
    Randomly kills applications
    Vader
    PYTHON
    PYTHON
    PYTHON

    View full-size slide



  78. Operator
    Datacenter
    RUBY GOLANG
    GOLANG
    GOLANG
    GOLANG
    NODE
    Skywalker Leia Solo
    RUBY
    VADER
    LEIA
    SOLO
    192.168.1.4
    192.168.1.9
    192.168.1.7
    192.168.1.253
    88:45:13:B6:87:C4
    94:CE:4F:C8:54:C3
    CA:9A:3D:7F:8B:CB
    72:30:9C:0D:1E:74
    Rebuilt on 04/20/2016
    Vader
    PYTHON
    PYTHON
    PYTHON

    View full-size slide



  79. Operator
    Datacenter
    RUBY GOLANG
    GOLANG
    GOLANG
    GOLANG
    NODE
    Skywalker Leia Solo
    RUBY
    VADER
    LEIA
    SOLO
    192.168.1.4
    192.168.1.9
    192.168.1.7
    192.168.1.253
    88:45:13:B6:87:C4
    94:CE:4F:C8:54:C3
    CA:9A:3D:7F:8B:CB
    72:30:9C:0D:1E:74
    Rebuilt on 04/20/2016
    Vader
    PYTHON
    PYTHON
    PYTHON

    View full-size slide

  80. THIS DOES NOT SCALE
    http://www.public-domain-image.com/free-images/objects/spring-scale-used-to-determine-ones-weight-in-pounds.jpg

    View full-size slide

  81. Type Work Resources
    CPU Scheduler Threads Physical Cores
    EC2 / Nova Virtual Machines Hypervisors
    Hadoop YARN MapReduce Jobs Client Nodes
    Cluster Scheduler Applications Machines
    https://upload.wikimedia.org/wikipedia/commons/a/a0/Cedar_Mtn_Wilderness.jpg

    View full-size slide

  82. http://i.huffpost.com/gen/1821469/images/o-LANGUAGES-facebook.jpg
    ANOTHER LIVE DEMO

    View full-size slide

  83. Higher Resource Utilization
    Decouple Work from Resources
    Better Quality of Service
    ADVANTAGES
    https://upload.wikimedia.org/wikipedia/commons/3/38/GiantEagleStowOhio.jpg

    View full-size slide

  84. Higher Resource Utilization
    ADVANTAGES
    https://upload.wikimedia.org/wikipedia/commons/3/38/GiantEagleStowOhio.jpg
    Bin Packing
    Over-Subscription
    Job Queueing

    View full-size slide

  85. Decouple Work from Resources
    ADVANTAGES
    https://upload.wikimedia.org/wikipedia/commons/3/38/GiantEagleStowOhio.jpg
    Abstraction
    API Contracts
    Standardization

    View full-size slide

  86. Better Quality of Service
    ADVANTAGES
    https://upload.wikimedia.org/wikipedia/commons/3/38/GiantEagleStowOhio.jpg
    Priorities
    Resource Isolation
    Pre-emption

    View full-size slide

  87. https://upload.wikimedia.org/wikipedia/commons/8/85/Trail_Ridge_Road_and_Longs_Peak_by_RO.jpg
    SCALABILITY
    BONUS

    View full-size slide

  88. Nomad
    Million Container
    Challenge
    1,000 Jobs
    1,000 Tasks per Job
    5,000 Hosts on GCE
    1,000,000 Containers

    View full-size slide

  89. http://i.huffpost.com/gen/1821469/images/o-LANGUAGES-facebook.jpg
    No one will ever need to schedule a million containers.

    View full-size slide

  90. https://upload.wikimedia.org/wikipedia/commons/thumb/e/e1/Bill_Gates,_WEF_2009_Davos.jpg/1280px-Bill_Gates,_WEF_2009_Davos.jpg
    640K ought to
    be enough for
    anybody

    View full-size slide

  91. https://upload.wikimedia.org/wikipedia/commons/d/d7/Wikimedia_Foundation_Servers-8055_24.jpg
    1. USE A CLOUD PROVIDER
    2. USE A SCHEDULER
    3. MEASURE EVERYTHING

    View full-size slide

  92. @SETHVARGO
      
    QUESTIONS?

    View full-size slide