Embracing the network - Smashing Conf

This version of the talk was originally presented at Smashing Conference in Barcelona, Spain on the 20th October 2015.

The network is intrinsically unreliable. More so, the network is out of your control as a developer. Therefore, we must design systems which embrace the unpredictability of the network and defend against it all costs. How can you prioritise the delivery of your core content? What best-practices can you use to optimise your assets? How can we design interfaces which adapt and respond to changing network conditions? And finally, how are new APIs such as ServiceWorker changing the way we think about the network?

During this talk Patrick will share his experiences delivering high-performance websites to millions of users over the past 3 years at The Guardian and Financial Times. Which – most importantly – are resilient to the network.

Transcript

1. Embracing the network
   Patrick Hamann — Smashing Conference, October 2015

   @patrickhamann
   Modern techniques for building resilient front ends
   !
   

   View Slide

2. View Slide

3. View Slide

4. Why?
   

   View Slide

5. https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing
   FT.com
   

   View Slide

6. View Slide

7. View Slide

8. View Slide

9. https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing
   1. The network is reliable.
   2. Latency is zero.
   3. Bandwidth is infinite.
   4. The network is secure.
   5. Topology doesn't change.
   6. There is one administrator.
   7. Transport cost is zero.
   8. The network is homogeneous.
   

   The fallacies of distributed computing 

   PETER DEUTSCH — SUN MICROSYSTEMS, 1994
   https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing
   

   View Slide

10. Multiple points of failure
    Dmytrii Shchadei — http://www.slideshare.net/metrofun/reduce-mobile-latency
    " # #
    #
    $
    Internal latency
    Firewalls, Load Balancers,
    Servers
    %
    Internet routing latency
    CDNs, ISPs, Caches, Proxies
    Control plane latency
    ~600ms on average UK 3G connection ~200ms
    

    View Slide

11. The radio state machine
    http://developer.android.com/training/efficient-downloads/efficient-network-access.html
    Radio standby Radio low power Radio full power
    2s latency
    1.5s latency
    Radio idle for
    5 seconds
    Radio idle for
    12 seconds
    

    View Slide

12. Title Text
    

    View Slide

13. Title Text
    Request Map: http://requestmap.webperf.tools/
    

    View Slide

14. Failure is inevitable
    • 1,000,000 page views a day
    • 3 years product life-time
    • (1000000*360)*3 = 1,080,000,000
    • 1 billion opportunities for something to go wrong.
    

    View Slide

15. Baking in the assumption that everything can and will fail
    leads you to think differently about how you solve problems.
    

    SAM NEWMAN — BUILDING MICROSERVICES, O’REILLY 2015
    http://shop.oreilly.com/product/0636920033158.do
    

    View Slide

16. 1. Testing and monitoring failure
    2. Designing for failure
    3. Embracing failure
    4. The future
    

    View Slide

17. Testing and monitoring failure
    

    View Slide

18. A single point of failure (SPOF) is a part of a system that, if it
    fails, will stop the entire system from working. SPOFs are
    undesirable in any system with a goal of high availability or
    reliability, be it a business practice, software application, or
    other industrial system.
    https://en.wikipedia.org/wiki/Single_point_of_failure
    

    View Slide

19. Front end SPOFs
    http://www.stevesouders.com/blog/2010/06/01/frontend-spof/
    Chrome Firefox IE Opera Safari
    External script Blank below Blank below Blank below Blank below Blank below
    Stylesheet Flash Flash Blank below Flash Blank below
    inlined 

    @font-face
    Delayed Flash Flash Flash Delayed
    Stylesheet
    @font-face
    Delayed Flash Totally blank Flash Delayed
    Script then

    @font-face
    Delayed Flash Totally blank Flash Delayed
    

    View Slide

20. http://uk.businessinsider.com/doubleclick-for-publishers-down-2014-11
    

    View Slide

21. http://uk.businessinsider.com/doubleclick-for-publishers-down-2014-11
    

    View Slide

22. http://uk.businessinsider.com/doubleclick-for-publishers-down-2014-11
    

    View Slide

23. View Slide

24. https://twitter.com/scottjehl/status/636303029533282304
    

    View Slide

25. https://en.wikipedia.org/wiki/Fallacies_of_distributed_computing
    

    View Slide

26. Comcast - https://github.com/tylertreat/Comcast
    $ comcast --device=eth0 --latency=250 --target-bw=1000 --packet-loss=10%
    

    View Slide

27. Jeremy Keith — https://flic.kr/p/o9thWy
    

    View Slide

28. Resource timing API
    http://www.w3.org/TR/resource-timing/
    var resourceList = window.performance.getEntriesByType(‘resource');
    for (i = 0; i < resourceList.length; i++) {
    if (/link|script/.test(resourceList[i].initiatorType)) {
    navigator.sendBeacon('https://stats.ft.com/', {
    resource: resourceList[i].name,
    timing: (resourceList[i].responseEnd - resourceList[i].startTime)
    });
    }
    }
    

    View Slide

29. View Slide

30. • Test 1st and 3rd party dependencies for SPOFs
    • Use network shaping to simulate mobile conditions
    • Monitor all response times, timeouts and errors
    • Alert if metrics pass their thresholds
    

    View Slide

31. 1. Testing and monitoring failure
    2. Designing for failure
    3. Embracing failure
    4. The future
    

    View Slide

32. Designing for failure
    

    View Slide

33. View Slide

34. There seems to be an excessive amount of waiting around for
    pages to refresh and load -it doesn't seem as quick as the
    previous version.
    Luke Wroblewski – http://www.lukew.com/ff/entry.asp?1797
    

    View Slide

35. View Slide

36. View Slide

37. View Slide

38. View Slide

39. Little big details — http://littlebigdetails.com/post/56782380031/pinterest-before-image-is-loaded-the-background
    

    View Slide

40. Time and perception
    Delay User perception
    0 - 100ms Instant
    100 - 300ms Small perceptible delay
    300 - 1000ms Machine is working
    1000+ ms Likely mental context switch
    10,000+ ms Task is abandoned
    

    View Slide

41. View Slide

42. View Slide

43. • Respond within a 1000ms
    • Avoid loading spinners
    • Consider using skeleton screens
    • Re-use familiar UI patterns within error messaging
    • Use human friendly copy, especially in error states
    

    View Slide

44. 1. Testing and monitoring failure
    2. Designing for failure
    3. Embracing failure
    4. The future
    

    View Slide

45. Embracing failure
    

    View Slide

46. Pre-connect
    Ilya Grigorik — Eliminating roundtrips wth pre-connect: https://www.igvita.com/2015/08/17/eliminating-roundtrips-with-preconnect/
    socket setup
    

    View Slide

47. Pre-connect
    
    
    function preconnectTo(url) {
    var hint = document.createElement("link");
    hint.rel = "preconnect";
    hint.href = url;
    document.head.appendChild(hint);
    }
    Ilya Grigorik — Eliminating roundtrips wth pre-connect: https://www.igvita.com/2015/08/17/eliminating-roundtrips-with-preconnect/
    

    View Slide

48. When should I use resource hints?
    Method Use case Examples Frequecy
    preconnect
    Initiating an early connection to
    critical resources
    Image hosts, CDNs and
    analytics
    Often
    prefetch Fetch individual resource
    CSS or JS required for next
    page
    Limited
    prerender Fetch and execute HTML page Next page in a checkout flow Use with care
    

    View Slide

49. View Slide

50. Caching
    Libraries Utilities Application
    Rarely change Frequent change
    

    View Slide

51. Caching
    Libraries Utilities Application
    Rarely change Frequent change
    

    View Slide

52. Caching
    Libraries Utilities Application
    Rarely change Frequent change
    &
    core.js
    &
    app.js
    

    View Slide

53. HTTP/2
    R.U.M data sampled over 1 month from https://next.ft.com/
    H2 Capable
    H2 Enabled
    Unsupported
    Mobile user load event
    0ms 2,000ms 4,000ms 6,000ms 8,000ms
    7,200
    5,325
    6,160
    

    View Slide

54. View Slide

55. Loading fonts
    https://tabatkins.github.io/specs/css-font-rendering/
    Chrome Opera Safari Firefox IE
    Timeout 3s 3s no timeout 3s 0
    Fallback Yes Yes n/a Yes Yes
    Swap Yes Yes n/a Yes Yes
    

    View Slide

56. CSS Font Rendering Controls Module
    https://tabatkins.github.io/specs/css-font-rendering/
    @font-face {
    font-family: 'MyShinyFont';
    src: url('http://fonts.ft.com/async.woff2');
    font-display: swap;
    }
    @font-face {
    font-family: 'MyBlockingFont';
    src: url: url('http://fonts.ft.com/blocking.woff2');
    font-display: block;
    }
    

    View Slide

57. '
    ServiceWorker
    & (
    %
    

    View Slide

58. '
    ServiceWorker
    & (
    %
    
    

    View Slide

59. '
    ServiceWorker
    & (
    %
    

    View Slide

60. Cache only
    self.addEventListener('install', function(event) {
    event.waitUntil(
    caches.open(cacheName).then(function(cache) {
    return cache.addAll([
    ‘http://fonts.ft.com/fonts.css',
    'http://fonts.ft.com/serif',
    'http://fonts.ft.com/sans-serif'
    ]);
    })
    );
    });
    https://jakearchibald.com/2014/offline-cookbook
    

    View Slide

61. Cache only
    https://jakearchibald.com/2014/offline-cookbook
    self.addEventListener('fetch', function(event) {
    // If this is a request for font file
    if (/fonts\.ft\.com/.test(event.request)) {
    // If a match isn't found in the cache, the response
    // will look like a connection error
    event.respondWith(caches.match(event.request));
    }
    });
    

    View Slide

62. X
    Cache only
    '
    & (
    %
    
    (
    X
    (
    (
    &
    

    View Slide

63. Loading CSS
    https://developers.google.com/web/fundamentals/performance/critical-rendering-path/analyzing-crp?hl=en
    Request page
    Network
    Render
    GET html
    Build DOM
    response
    Build CSSOM Render page
    GET css
    GET js
    response
    response
    idle idle
    Render blocking
    Run JS
    Render blocking
    

    View Slide

64. Loading CSS
    Request page
    Network
    Render
    GET html
    Build DOM
    response
    Build CSSOM Render page
    GET css
    GET js
    response
    response
    idle idle
    Async
    Render blocking
    Run JS
    https://developers.google.com/web/fundamentals/performance/critical-rendering-path/analyzing-crp?hl=en
    

    View Slide

65. https://github.com/filamentgroup/loadCSS
    

    View Slide

66. Loading CSS
    Request page
    Network
    Render
    GET html
    Build DOM
    response
    Build CSSOM Render page
    GET css response
    idle idle
    Async
    https://developers.google.com/web/fundamentals/performance/critical-rendering-path/analyzing-crp?hl=en
    Reflow
    

    View Slide

67. Stale-while-revalidate
    self.addEventListener('fetch', function(event) {
    // Get stale from cache
    event.respondWith(
    caches.open(cacheName).then(function(cache) {
    return cache.match(event.request).then(function(response) {
    // If in cache relvalidate
    var network = fetch(event.request).then(function(response){
    if(response.status < 400) {
    cache.put(event.request, response.clone());
    }
    });
    // Return stale or network if none
    return response || network;
    });
    })
    );
    });
    https://jakearchibald.com/2014/offline-cookbook
    

    View Slide

68. Stale-while-revalidate
    '
    & (
    %
    
    (
    

    View Slide

69. Stale-if-error
    self.addEventListener('fetch', function(event) {
    // Always fetch response from the network
    event.respondWith(
    fetch(event.request).then(function(response) {
    return caches.open(cacheName).then(function(cache) {
    // If we received an error response
    if(!response.ok) {
    return cache.match(event.request)
    } else {
    // Response was healthy so update cached version
    cache.put(event.request, response.clone());
    return response;
    }
    });
    })
    );
    });
    

    View Slide

70. Timeouts
    '
    & (
    %
    X *
    X
    

    View Slide

71. Timeout race
    function timeout(delay) {
    return new Promise(function(resolve, reject) {
    setTimeout(function() {
    resolve(new Response('', {
    status: 408,
    statusText: 'Request timed out.'
    }));
    }, delay);
    });
    }
    self.addEventListener('fetch', function(event) {
    // Attempt to fetch with timeout
    event.respondWith(Promise.race([timeout(2000), fetch(event.request)]));
    });
    

    View Slide

72. Dead letter queue
    '
    & (
    %
    X
    +
    offline
    

    View Slide

73. Dead letter queue
    '
    & (
    %
    +
    Online
    

    View Slide

74. Dead letter queue
    var queue = {};
    function queueFailedRequest(request) {
    queue[Date.now()] = request;
    }
    self.addEventListener('fetch', function(event) {
    if(/track\.ft\.com/.test(event.request.url) {
    event.respondWith(
    fetch(event.request).then(function(response){
    if(response.status >= 500) {
    return Response.error();
    } else {
    return response;
    }
    }).catch(function() {
    queueFailedRequest(event.request);
    })
    );
    }
    });
    

    View Slide

75. Dead letter queue
    self.addEventListener('sync', function(event) {
    event.waitUntil(
    return Promise.all(queue.map(function(request) {
    return fetch(request)
    }))
    );
    });
    

    View Slide

76. • Pre-connect sockets
    • Pre-fetch and cache critical resources
    • Bundle resources by their frequency of change
    • Serve stale and revalidate
    • Serve stale if error
    • Always use timeouts
    • Treat the network as an enhancement!
    

    View Slide

77. 1. Testing and monitoring failure
    2. Designing for failure
    3. Embracing failure
    4. The future
    

    View Slide

78. The future
    

    View Slide

79. View Slide

80. Offline
    Network unavaliable
    

    View Slide

81. Push notifications
    

    View Slide

82. Background sync
    navigator.serviceWorker.ready.then(function(registration) {
    regitration.periodicSync.register({
    tag: 'get-latet-news', // default: ''
    minPeriod: 12 * 60 * 60 * 1000, // default: 0
    powertState: 'avoid-draining', // default: 'auto'
    networkState: 'avoid-cellular' // default: 'online'
    }).then(function(periodicsyncReg) {
    // success
    }, function() {
    // failure
    });
    });
    

    View Slide

83. 1. Testing and monitoring failure
    2. Designing for failure
    3. Embracing failure
    4. The future
    

    View Slide

84. Conclusion
    

    View Slide

85. Thanks
    @patrickhamann
    speakerdeck.com/patrickhamann

    github.com/Financial-Times
    !
    ,
    -
    Do you want to help build this stuff? Join in.
    jobs@labs.ft.com
    

    View Slide