Necessary Performance Measures: npm at Scale

Talk on caching and distribution at MelbJS meetup in Melbourne, Australia

Transcript

1. Necessary Performance
   Measures: at Scale
   Brought to you by @rockbot
   

   View Slide

2. Package Management for
   Node.js
   • The Open Source Dream
   
   !
   • ~70k packages
   
   • up to 10 million downloads/
   day
   

   View Slide

3. 221,845,731)
   !10,000!!
   !20,000!!
   !30,000!!
   !40,000!!
   !50,000!!
   !60,000!!
   !70,000!!
   !80,000!!
   !90,000!!
   !-!!!!
   !10,000,000!!
   !20,000,000!!
   !30,000,000!!
   !40,000,000!!
   !50,000,000!!
   !60,000,000!!
   !70,000,000!!
   !80,000,000!!
   !90,000,000!!
   !100,000,000!!
   !110,000,000!!
   !120,000,000!!
   !130,000,000!!
   !140,000,000!!
   !150,000,000!!
   !160,000,000!!
   !170,000,000!!
   !180,000,000!!
   !190,000,000!!
   !200,000,000!!
   !210,000,000!!
   !220,000,000!!
   !230,000,000!!
   !240,000,000!!
   8/29/12! 12/7/12! 3/17/13! 6/25/13! 10/3/13! 1/11/14! 4/21/14!
   Downloads)of)npm)packages)in)previous)month))
   &)Total)number)of)npm)Packages)of)node.js)code)
   !!!Downloads!of!npm!packages!in!previous!month!
   !!!Total!Number!of!npm!Packages!
   

   View Slide

4. Performance = Perception
   • Best request vs worst request
   
   • Focus on the few, super-slow instances: improve
   performance like whoa
   
   !
   • Caching vs Distribution
   

   View Slide

5. Caching
   

   View Slide

6. Data Layer
   Query Caching
   Database-level
   performance
   enhancements
   !
   innoDB buffer pool,
   CouchDB’s ACID
   architecture
   

   View Slide

7. Query Layer
   Save CPU
   Preprocessed/aggregated
   data; scheduled handling
   of data
   !
   CouchDB views,
   Download Counts
   

   View Slide

8. App Layer
   Per-User
   Data that doesn’t change
   per user
   !
   Storing user session data
   in Redis
   

   View Slide

9. Edge Server
   Layer
   Centralized/Global
   CDN, static assets
   !
   Fastly
   

   View Slide

10. Client Layer
    Distributed
    HTTP headers, local
    storage, etc.
    !
    Browser caching
    

    View Slide

11. Distribution
    

    View Slide

12. Split Load
    More boxes,
    more specialized boxes
    read-only, write-only
    !
    One process per box
    Redis, CouchDB,
    ElasticSearch
    

    View Slide

13. Add
    Resiliency
    Multi-cloud, multi-zone
    East/West,
    multiple datacenters
    !
    Fallback to least effort/
    most common usage
    i.e. read-only
    

    View Slide

14. Boost with
    Third-Party
    Services
    Global content
    distribution network
    CDN
    

    View Slide

15. The End?
    

    View Slide

16. Thank You!
    !
    Raquel Vélez
    @rockbot
    raquel@ js.com
    

    View Slide