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

Scalability vs. Performance

Scalability vs. Performance

Creating scalable web sites that can handle many simultaneous requests and still provide fast experience for each user is hard. Historically, the industry was not differentiating Scalability and Performance, but with emergence of front-end engineering, new field of Web Performance Optimization was born and it became critical to approach them separately.

Sergey Chernyshev (@sergeyche) compares these two directions of web engineering and describes the differences between them. He also covers current performance trends and describes different approaches to take in order to measure and analyze Web Performance in comparison to traditional methods that are successfully used to test scalability of web systems.

Talk given at New York Web Performance Meetup on March 27, 2013
http://www.meetup.com/Web-Performance-NY/events/101100592/

Sergey Chernyshev

March 27, 2013
Tweet

More Decks by Sergey Chernyshev

Other Decks in Technology

Transcript

  1. Size & Speed
    Scalability vs. Performance
    NY Web Performance Meetup
    March 27, 2013

    View full-size slide

  2. About
    • Doing web dev and
    operations since 1996
    • Tech Director, Web Ops
    and Dev @ truTV/Turner
    • Creator of ShowSlow
    • Organizer at NY WebPerf
    • Board member of WPO
    Foundation

    View full-size slide

  3. Why Do We Scale?

    View full-size slide

  4. Available Resources
    CPU Memory
    Disk Network

    View full-size slide

  5. CPU & Memory
    • Classical scale problem
    • Scale up (Moor's law)
    • Scale out (cloud)

    View full-size slide

  6. Storage
    • More available
    • Speed and reliability
    • Slower then in the past / SSD

    View full-size slide

  7. Network ???
    • Easy to scale for static assets
    • Trouble if you hit dynamic

    View full-size slide

  8. User Experience

    View full-size slide

  9. How Page Loads
    ~8 sec above the fold
    ~0.5 sec for first request
    only 6.25% on back end
    12sec, 157 req, ~1Mb
    http://bit.ly/pageload-video
    http://bit.ly/pageload-filmstrip

    View full-size slide

  10. Load Testing
    apply load,
    measure backend
    performance degradation

    View full-size slide

  11. "Synthetic" testing
    time ALL components of page load
    HAR

    View full-size slide

  12. Real User Measurement
    measure key performance metrics for REAL users
    Boomerang.js

    View full-size slide

  13. Speed, is it important?
    RUM data. Aaron Kulick, Cliff Crocker @ Wallmart Labs

    View full-size slide

  14. Speed, it is important!
    RUM data. Aaron Kulick, Cliff Crocker @ Wallmart Labs

    View full-size slide

  15. Speed it up
    for one user!

    View full-size slide

  16. New York
    Web Performance
    Meetup
    Sergey Chernyshev

    @sergeyche
    @perfplanet

    View full-size slide