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

Building in-house CDN by Fredrik Widlund from SVT

Building in-house CDN by Fredrik Widlund from SVT

Fredrik Widlund from SVT sharing their experiences from building their own in-house CDN. Presented at Streaming Tech Sweden on Nov 22, 2017

Streaming Tech Sweden

November 22, 2017
Tweet

More Decks by Streaming Tech Sweden

Other Decks in Technology

Transcript

  1. whoami • Programmer / Systems architect • Like to take

    things apart • Like to build on ideas and concepts • Security (Defcom) • Streaming (Qbrick, SVT) • AI / Computer Vision (OculusAI) • Big data analytics (Meltwater) • https://github.com/fredrikwidlund
  2. 2010 • SVT uses Qbrick CDN for streaming • SVT

    is Qbrick’s largest customer • Losing that contract is a business risk
  3. 2010 • Could SVT build a CDN of their own

    + Content + Volume - Domain knowledge - Complexity
  4. 2015 • Qbrick lost the SVT contract to Akamai •

    SVT second in Sweden in terms of volume • Users online are about 6 times more expensive • Rapid cost growth • Innovation blockage
  5. Business proposition • Put together a team of 4-5 people

    • Full scale production in one year • Distribute 30% of content ourselves • Self-financing • Break trend with rapid cost increase • Address innovation issues
  6. TRAFFIC NETWORK DISTRIBUTION PACKAGING PROVIDERS STORAGE General technology stack •

    Commodity hardware • Stable chipsets (Intel) • Persistence (Disk, Flash, RAM) • Networking (NICs, switches, routers) • Linux (CoreOS) • systemd • Docker
  7. TRAFFIC NETWORK DISTRIBUTION - NGINX / NVMe PACKAGING - WOWZA

    PROVIDERS STORAGE - SAS / JBOD / ZFS / GLUSTER
  8. TRAFFIC DISTRIBUTION - NGINX / NVMe PACKAGING - WOWZA PROVIDERS

    STORAGE - SAS / JBOD / ZFS / GLUSTER NETWORK
  9. TRAFFIC - TRANSIT / PUBLIC PEERING / PRIVATE PEERING DISTRIBUTION

    - NGINX / NVMe PACKAGING - WOWZA PROVIDERS STORAGE - SAS / JBOD / ZFS / GLUSTER NETWORK
  10. Other things we worked on • Load balancing • Redundancy

    • API performance • Alerting and error handling • Quality measurements • Event and log management • Provisioning • HTTPS
  11. Business value • In full production after a year •

    First year we handled 30% of the total volume in-house • Currently we handle about 50% • Distribution costs significantly lowered • Project was self-financed • Innovation boost • Knowledge boost
  12. Takeaways • Simplify when possible • Optimise when needed •

    Keep focus on business value • Don’t start from scratch