Go for speed

A look at a use case for Go at Iconfinder, which shaved ~40 ms off asset load times.

Transcript

1. Go for speed
   

   View Slide

2. Nick Bruun
   github.com/nickbruun
   @nickbruun
   bruun.co
   CTO @ Iconfinder
   Fairly carnivorous
   

   View Slide

3. View Slide

4. View Slide

5. View Slide

6. Trouble in icon paradise
   

   View Slide

7. View Slide

8. View Slide

9. View Slide

10. View Slide

11. 41 ms
    average response time from Django application
    during peak hours
    

    View Slide

12. ~2,000 ms
    wasted per search
    

    View Slide

13. Perceivably slower
    

    View Slide

14. The internals
    

    View Slide

15. View Slide

16. Icon ID
    Icon version ID
    Raster size
    https://www.iconfinder.com/icons/298255/677266/128/raster
    https://cdn0.iconfinder.com/data/icons/seo-smart-pack/128/
    grey_new_seo2-43-128.png
    Public raster URL
    Protected raster URL
    

    View Slide

17. View Slide

18. Constant
    

    View Slide

19. Signing requests
    

    View Slide

20. https://www.iconfinder.com/icons/298255/677266/128/raster?token=…
    

    View Slide

21. View Slide

22. we know if the user has access right now
    When returning the link,
    

    View Slide

23. we know if the user has access right now
    When returning the link,
    

    View Slide

24. We’re okay with “right now”
    being an approximation
    

    View Slide

25. Time limited token
    

    View Slide

26. = "---"
    

    View Slide

27. = "---"
    = hmac_sha1(, )
    

    View Slide

28. = "---"
    = hmac_sha1(, )
    = "--base64()"
    

    View Slide

29. But, who checks the token?
    

    View Slide

30. Enter Go
    The World’s least Google-able language… from Google
    

    View Slide

31. View Slide

32. Why Go?
    

    View Slide

33. Networked service
    

    View Slide

34. Highly concurrent
    ~3,000 requests/second from 100 connections/core
    

    View Slide

35. Low memory footprint
    14.09 MiB after 21 days
    

    View Slide

36. Low CPU usage
    

    View Slide

37. Static binary
    Cross-compilation = deployment heaven
    

    View Slide

38. Less decisions to make
    

    View Slide

39. Caveats
    

    View Slide

40. Structured logging
    

    View Slide

41. DIY
    

    View Slide

42. Mature runtime
    !=
    Mature language
    

    View Slide

43. Conciseness
    608 lines of code
    

    View Slide

44. The end result
    

    View Slide

45. 1.2 ms
    average response time at peak hours
    

    View Slide

46. 0.1 %
    CPU usage at peak hours
    

    View Slide

47. Go is awesome
    

    View Slide

48. Go is awesome for simple services
    

    View Slide

49. Go is a trade-off
    

    View Slide

50. Go is a trade-off – choose wisely
    

    View Slide

51. return
    

    View Slide