High Performance Web Applications

Building high performance web applications on AWS.

Transcript

1. High Performance
   Web Applications
   matthew@amazon.com
   Dr. Matt Wood
   @mza
   

   View Slide

2. Hello
   

   View Slide

3. Building blocks for wonderful applications
   Storage, compute, databases, managed services
   

   View Slide

4. View Slide

5. Retail Merchant
   services
   Web
   services
   

   View Slide

6. Retail Merchant
   services
   Web
   services
   

   View Slide

7. Getting started with
   Amazon Web Services
   quick review
   H
   

   View Slide

8. Utility computing
   Available at the end of an API call.
   Metered billing.
   H
   

   View Slide

9. Family of 20 services
   Adding new services and refinement weekly.
   Rapid rate of innovation.
   H
   

   View Slide

10. Undi!erentiated heavy lifting
    Focus on your app, customers and business
    H
    

    View Slide

11. H
    

    View Slide

12. aws.amazon.com
    H
    Free account.
    Free tier.
    

    View Slide

13. Building High Performance
    Web Applications
    r
    

    View Slide

14. Deliver a great experience to
    your customers.
    Responsive, fast web applications.
    r
    

    View Slide

15. g DECOUPLED
    SCALE
    OUT
    S
    OPTIMIZE COST
    FOR
    C
    AUTOMATE
    v CALIBRATED FOR:
    CPU performance v CALIBRATED FOR:
    IO performance
    5patterns for
    performance
    

    View Slide

16. g DECOUPLED
    

    View Slide

17. Monolithic Mega System v2
    

    View Slide

18. Application
    Assets
    Analytics
    

    View Slide

19. Application
    Assets
    Analytics
    

    View Slide

20. Images, javascript, stylesheets,
    HTML pages
    Deliver as fast as possible to customers
    r
    

    View Slide

21. Websites without the servers
    Highly durable: 99.999999999%
    Zero config: fire and forget
    r
    

    View Slide

22. View Slide

23. Low latency delivery of assets
    Local caching for fast downloads
    Amazon CloudFront
    r
    

    View Slide

24. P
    P
    

    View Slide

25. Create content distributions
    From objects in S3
    Any custom origin
    Dynamic, streaming and static content
    r
    

    View Slide

26. P
    P
    P
    

    View Slide

27. CNAME?
    Map to any custom URL
    assets.domain.com, www.domain.com
    r
    

    View Slide

28. Application
    Assets
    Analytics
    

    View Slide

29. Application
    Assets
    Analytics
    

    View Slide

30. Application architecture
    Small things, loosely coupled
    r
    

    View Slide

31. Application
    Assets
    Analytics
    

    View Slide

32. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

33. Shared responsibility
    Infrastructure by AWS
    OS, application and data by customers
    Tools to help meet best practices
    r
    

    View Slide

34. Security groups
    Customer controlled firewalls
    All ports are closed by default
    r
    

    View Slide

35. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

36. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    x
    

    View Slide

37. Decoupling provides an architectural
    framework for performance
    Easier to separate concerns: assets, streaming, etc.
    Easier to manage and scale.
    r
    

    View Slide

38. SCALE
    OUT
    

    View Slide

39. Elasticity is a tenet of the cloud
    Scale up when you need to
    Scale down when you don’t
    r
    

    View Slide

40. Predictable peaks
    r
    Time
    Demand
    Capacity
    

    View Slide

41. Predictable peaks
    r
    Time
    Demand
    Capacity
    

    View Slide

42. Variable usage
    r
    Time
    Demand
    Capacity
    

    View Slide

43. Variable usage
    r
    Time
    Demand
    Capacity
    

    View Slide

44. Variable usage
    r
    Time
    Demand
    

    View Slide

45. Horizontal scaling
    drives performance up
    ...and costs down
    r
    

    View Slide

46. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

47. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

48. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

49. Amazon Machine Images
    encapsulate your app
    Customize your virtual machines
    AMI can be the unit of deployment
    r
    

    View Slide

50. 1. Spin up a new instance
    using your AMI
    Configured to bootstrap the app.
    r
    

    View Slide

51. 2. Pull down latest code base
    From S3 or code repository
    r
    

    View Slide

52. 3. Start up the app on the instance
    Ready to receive requests
    r
    

    View Slide

53. 4. Register with the load balancer
    Job done.
    r
    

    View Slide

54. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

55. Horizontal scaling allows for
    fault tolerance
    r
    

    View Slide

56. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

57. Load balancer
    App servers
    Database
    Asynchronous processing
    :(
    Application
    Assets
    Analytics
    

    View Slide

58. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

59. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

60. Deploying and scaling a datastore
    r
    Follow the same patterns:
    horizontal scale, availability, automation.
    

    View Slide

61. One question:
    r
    Does your application require a
    strict, controlled schema for query flexibility?
    

    View Slide

62. Amazon Relational Database Service
    r
    Management systems: CRM, ERP, finance
    Does your application require a strict, controlled schema
    for query flexibility? Yes
    

    View Slide

63. Focus on your app
    r
    Handles tedious database admin tasks
    Designed for availability
    

    View Slide

64. Multi-engine
    r
    MySQL, Oracle, Microsoft SQL Server
    Up and running in six clicks
    

    View Slide

65. Point in time snapshots
    r
    Automatic. Easy recovery.
    

    View Slide

66. High availability
    r
    Deployed across multiple availability zones.
    Synchronous writes.
    

    View Slide

67. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

68. Read replicas
    r
    Perfect for read heavy applications.
    Asynchronous writes.
    Replication metrics available.
    

    View Slide

69. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

70. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

71. Amazon DynamoDB
    r
    Web apps, social apps, mobile apps,
    user generated content,
    unstructured data integration, lots of data.
    Does your application require a strict, controlled schema
    for query flexibility? No
    

    View Slide

72. Focus on your app
    r
    Managed NoSQL database service.
    No schema.
    

    View Slide

73. Unlimited scale
    r
    Unlimited storage
    Pay as you go
    

    View Slide

74. High performance
    r
    Single digit millisecond latencies
    

    View Slide

75. Zero admin
    r
    No instances to manage
    Tiny API, perfectly formed
    

    View Slide

76. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

77. Asynchronous processing
    r
    Resize images. Transcode movies.
    Resample media.
    

    View Slide

78. Follow the same best practices
    r
    Decouple. Stateless. Horizontal scale.
    

    View Slide

79. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

80. CloudFront for dynamic content
    r
    Edge caching for dynamic content
    Cache by query string parameters
    Multiple origin servers
    Persistent connections to origin servers
    

    View Slide

81. C
    AUTOMATE
    

    View Slide

82. Automate everything
    The tedious.
    The time consuming.
    The error prone.
    r
    

    View Slide

83. For example:
    when increasing capacity
    Launch instance.
    Update application code.
    Register with load balancer.
    r
    

    View Slide

84. Auto-scaling
    CloudWatch
    Auto-scaling service
    Elastic Load Balancer
    r
    

    View Slide

85. Set operational thresholds
    Network capacity > 80%
    Memory utilization > 75%
    CPU < 50%
    r
    

    View Slide

86. Automatically respond to
    operational alarms
    r
    Network capacity > 80%: add instances
    Memory utilization > 75%: add instances
    CPU < 50%: remove instances
    

    View Slide

87. Set bounds
    Step change configured by number or
    percentage of instances
    r
    

    View Slide

88. Custom metrics
    Respond to custom metrics
    Page load time
    Database query time
    r
    

    View Slide

89. Automatic load balancer
    configuration
    Instances added/removed automatically
    Health check for running application before
    routing tra"c
    r
    

    View Slide

90. Stateless operations are important
    Decoupled, but stateless
    Instances are naive to other instances
    r
    

    View Slide

91. Treat your data as a royal garden
    Backup to S3: high durability
    Compute is a disposable resource
    r
    

    View Slide

92. v CALIBRATED FOR:
    CPU performance
    

    View Slide

93. Range of resources
    Choose the best instance type
    for performance
    r
    

    View Slide

94. r
    t1.micro m1.small cc2.8xlarge
    

    View Slide

95. r
    t1.micro m1.small cc2.8xlarge
    High memory
    High CPU
    High CPU + memory
    

    View Slide

96. Load balancer
    App servers
    Database
    Asynchronous processing
    Application
    Assets
    Analytics
    

    View Slide

97. 1 instance for 100 hours
    =
    100 instances for 1 hour
    

    View Slide

98. Balance cost/performance
    Mix instance types for application servers,
    databases, ancillary services, analytics etc.
    r
    

    View Slide

99. Migrate to large instance types
    As your application grows
    Use 64 bit from the outset, if possible
    r
    

    View Slide

100. v CALIBRATED FOR:
     IO performance
     

     View Slide

101. IO bound web applications are
     more common
     Application requires predictable, consistent IO
     r
     

     View Slide

102. Provisioned throughput with
     DynamoDB
     No need to provision storage
     Provision IO requirements:
     reads and writes per second
     r
     

     View Slide

103. View Slide

104. View Slide

105. View Slide

106. View Slide

107. Scale up, and carry on
     DynamoDB will manage resources to
     achieve and maintain throughput
     r
     

     View Slide

108. Low latency, unlimited scale
     Writes acknowledged when saved to disk in
     multiple data centres. Backed on SSDs.
     r
     

     View Slide

109. ...and SSDs for all
     

     View Slide

110. hi1.4xlarge instances
     High capacity, high performance storage
     2 x 1Tb SSD drives
     r
     

     View Slide

111. Perfect match for databases
     Relational and NoSQL storage
     r
     

     View Slide

112. Perfect match for databases
     Relational and NoSQL storage
     90k to 120k IOPS
     r
     

     View Slide

113. High performance network
     10 gigabit ethernet
     Placement groups
     Perfect for distributed stores
     r
     

     View Slide

114. S
     OPTIMIZE COST
     FOR
     

     View Slide

115. Economies of scale
     Drive down prices for customers
     20 price drops in past 6 years
     r
     

     View Slide

116. Maximize utilization
     Greater utilization drives down cost
     r
     

     View Slide

117. Reserved capacity
     Time
     Utilization
     

     View Slide

118. Reserved instances
     Small upfront payment
     Guaranteed capacity for 1 or 3 years
     Lower hourly rate
     r
     

     View Slide

119. Reserved instance resale
     Sell your reservations on the resale market
     Sellers set your own price
     Buyers select and check out
     r
     

     View Slide

120. Reserved capacity
     Time
     Utilization
     

     View Slide

121. Reserved capacity
     On-demand
     Time
     Utilization
     

     View Slide

122. Reserved capacity
     On-demand
     Time
     Utilization
     

     View Slide

123. Spot instances
     Name your price computing
     r
     

     View Slide

124. Spot instances
     Lower costs
     ‘Overclock’ your app: more bang for your buck
     r
     

     View Slide

125. Load balancer
     App servers
     Database on SSDs
     Asynchronous processing
     Application
     Assets
     Analytics
     

     View Slide

126. Load balancer
     App servers
     Database on SSDs
     Asynchronous processing
     Application
     Assets
     Analytics
     

     View Slide

127. Load balancer
     App servers
     Database on SSDs
     Asynchronous processing
     Application
     Assets
     Analytics
     

     View Slide

128. Become price aware
     Price is a driver of your architecture
     r
     

     View Slide

129. Automated bidding
     Automate bids based on on-demand price
     Run on spot only when cost e!ective
     r
     

     View Slide

130. Continuous cost optimization
     Evaluate your reserved instance/on-demand/
     spot mix regularly as app requirements change
     r
     

     View Slide

131. 5patterns for
     performance
     g DECOUPLED
     

     View Slide

132. 5patterns for
     performance
     g DECOUPLED
     SCALE
     OUT
     

     View Slide

133. 5patterns for
     performance
     g DECOUPLED
     SCALE
     OUT
     S
     OPTIMIZE COST
     FOR
     

     View Slide

134. 5patterns for
     performance
     g DECOUPLED
     SCALE
     OUT
     S
     OPTIMIZE COST
     FOR
     v CALIBRATED FOR:
     CPU performance v CALIBRATED FOR:
     IO performance
     

     View Slide

135. 5patterns for
     performance
     g DECOUPLED
     SCALE
     OUT
     S
     OPTIMIZE COST
     FOR
     v CALIBRATED FOR:
     CPU performance v CALIBRATED FOR:
     IO performance
     C
     AUTOMATE
     

     View Slide

136. Thank you
     aws.amazon.com
     @mza
     matthew@amazon.com
     

     View Slide