Brace for impact

How to prepare for high load scenarios and traffic spikes - Fight the "Slashdot Effect". Held at @bephpug, Berlin.

Transcript

1. @bephpug 2012
   Christoph Lühr
   @chluehr
   Brace for impact
   

   View Slide

2. basilicom
   

   View Slide

3. View Slide

4. High Load Scenarios
   (short term)
   

   View Slide

5. §
   DISCLAIMER
   PHP
   

   View Slide

6. ( if successful )
   ==
   TV
   

   View Slide

7. /.
   "Slashdot Effect"
   ( now: Facebook, Bild.de, ... )
   

   View Slide

8. Bäm!
   

   View Slide

9. Why care?
   

   View Slide

10. $$$
    TV ads are expensive
    downtime == wasted money
    

    View Slide

11. Why me?
    5 yrs++
    Deutsche Bahn, Ferrero, Peugeot
    BMW, Sony, UMG, studiVZ, ...
    

    View Slide

12. Image source: http://www.flickr.com/photos/stinajonsson/3932774410 CC BY-NC 2.0
    

    View Slide

13. Flight Plan
    1. MAKE A PLAN
    2. MAKE IT SCALE?
    3. MAKE IT FAST!
    4. [ TEST, SIZE ]
    5. WATCH IT!
    6. BONUS: EXAMPLES
    } Architecture
    

    View Slide

14. Brace for impact
    

    View Slide

15. Brace for impact
    Plan
    

    View Slide

16. Knowledge is Power
    Media plans
    TV schedules
    

    View Slide

17. View Slide

18. Planning
    ● just Peaks matter
    ● Media / marketing exposure?
    ● Target group? Click Rate?
    ● Max. daily visits? Top hour?
    ● Pages ^ visit, resources ^ page?
    ● Do not forget storage / traffic requirements.
    

    View Slide

19. Planning
    

    View Slide

20. Planning
    

    View Slide

21. Architecture
    ( & Tool Time )
    

    View Slide

22. E l a s t i c i t y
    Scaling
    Fancy Solution
    

    View Slide

23. 1 4 5
    7 n
    1 2 3
    6
    "just add more boxes.."
    

    View Slide

24. really?
    

    View Slide

25. System complexity ++
    

    View Slide

26. Database?
    

    View Slide

27. Time to spawn?
    .......
    

    View Slide

28. TV-Ad
    10 sec. later
    10 min. later
    

    View Slide

29. Performance ++
    Traditional Solution
    

    View Slide

30. Performance
    SERVER BROWSER USER
    

    View Slide

31. Performance
    SERVER BROWSER USER
    FAST!
    

    View Slide

32. Performance Rules
    

    View Slide

33. Rule #1
    Don't get hit.
    ● landing page on a different server.
    

    View Slide

34. Rule #2
    Let others do the work
    ● use a Content Delivery Network
    (CloudFlare is free).
    

    View Slide

35. Rule #3
    Serve static content
    

    View Slide

36. SERVE
    STATIC
    CONTENT!
    

    View Slide

37. Rule #3
    Serve static content
    ● use a caching reverse proxy:
    Varnish
    ● NGINX is fast, too
    ● "static" can even mean only 5 mins.
    

    View Slide

38. Rule #4
    Serve static content once
    ● use correct cache headers
    ● leverage browser caches
    ● allow external proxies to cache
    your content.
    

    View Slide

39. Rule #5
    Minimize requests
    ● compress HTML
    ● combine / minify JS, CSS
    ● inline images
    ● == get a good SpeedRanking.
    

    View Slide

40. Rule #6
    Avoid dynamic requests
    ● load data upfront
    ● stay in browser, validate via JS first
    ● collect data / batch requests
    ● load only on action (captcha, ...).
    

    View Slide

41. Rule #7
    small
    is BIG
    ● don't load a framework for an insert()
    ● just write a logfile?
    ● use nodeJS.
    

    View Slide

42. Rule #8
    De-couple / postpone
    ● use queuing and finite # of workers
    ● use external services
    (transloadit, youtube, ...).
    

    View Slide

43. Rule #9
    Degrade gracefully
    ● skip doing hard work (thumbnails,...)
    ● hide expensive functionality (edit)
    ● ... to keep the site online.
    

    View Slide

44. Secret Rule #10
    Cheat
    ● discard requests
    ● stats & voting: sample only 10%
    ● simulate server calls (spinner).
    

    View Slide

45. Final Rule #11
    Don't go offline.
    ● put campaign on different server
    ● limit resources, don't kill your server
    ● if all fails, be sure to display a notice.
    

    View Slide

46. View Slide

47. Testing
    ● Goal: Determine app limits/threshold
    ● Testing is HARD
    ● Test the production environment
    ● Monitor performance & resources
    ● httperf, apache bench, siege, JMeter.
    

    View Slide

48. Sizing & Tuning
    

    View Slide

49. Sizing & Tuning
    ● RAM
    ● CPU cores / CPU speed
    ● IO: Storage Type / Storage Size
    ● Cache memory (Varnish)
    ● MySQL tuning (Caches, ...)
    ● System/Network tuning (max_open_files, ...)
    ● Apache (max. childs, ...)
    

    View Slide

50. Rinse, repeat.
    (building, testing, sizing)
    

    View Slide

51. Online:
    Monitoring
    Image source: http://www.officialpsds.com/Green-Radar-Screen-PSD61002.html Attribution
    

    View Slide

52. Monitoring
    ● Access (Google Analytics)
    ● External Availability (Pingdom, ...)
    ● Speed, Latency &
    Resources (New Relic, ...)
    ● Application (# participants, ... StatHat)
    ● OR homegrown (statsd, etc.).
    

    View Slide

53. Examples
    

    View Slide

54. Forms
    

    View Slide

55. Forms
    1
    2
    3
    

    View Slide

56. Gallery detail pages
    http://foo.com/view/?key=3Jsk31KOsk2
    IMAGE
    

    View Slide

57. Gallery detail pages
    http://foo.com/view/?key=3Jsk31KOsk2 ==> :-(
    http://foo.com/view/ #3Jsk31KOsk2 ==> :-)
    static!
    3Jsk31KOsk2.JPG
    

    View Slide

58. Simplified setup
    

    View Slide

59. Summary
    

    View Slide

60. SERVE
    STATIC
    CONTENT!
    

    View Slide

61. CRITICAL
    LOAD FOLLOWS
    

    View Slide

62. Suggestions?
    Questions?
    

    View Slide

63. Contact
    Christoph Lühr
    eMail: luehr@r-pentomino.de, christoph.luehr@basilicom.de
    Twitter: @chluehr
    Slides license
    Attribution-NonCommercial-ShareAlike 3.0
    http://creativecommons.org/licenses/by-nc-sa/3.0/
    Thanks
    ... and always have a safe landing!
    

    View Slide

64. Links
    ● Google PageSpeed Insights
    https://developers.google.com/speed/pagespeed/insights
    ● Transloadit: Image / Video uploads / conversion
    https://transloadit.com/
    ● Pingdom Monitoring & Tools
    http://tools.pingdom.com/fpt/
    ● New Relic application & server monitoring
    http://newrelic.com/
    ● CloudFlare CDN
    https://www.cloudflare.com/
    

    View Slide

65. Links
    ● Continuous Performance Testing
    http://qafoo.
    com/talks/12_05_ipc_se_continuous_performance_tests.pdf
    ● Varnish
    https://www.varnish-cache.org/
    ● Varnish Speed
    http://kly.
    no/posts/2010_10_23__High_End_Varnish___275_thousand_request
    s_per_second___.html
    

    View Slide

66. Links
    ● NGINX
    http://www.slideshare.net/Edorian/nginx-php-fpm-the-
    webserver-you-might-actually-like-php-usergroup-berlin
    ● High Performance Web Sites - Steve Souders
    http://shop.oreilly.com/product/9780596529307.do
    ● Pound Load Balancer, Reverse Proxy & SSL Offloader
    http://www.apsis.ch/pound
    

    View Slide