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

Infrastructure in LINE / LINE Campus Talk in Hong Kong by Yoshihiro Saegusa

Infrastructure in LINE / LINE Campus Talk in Hong Kong by Yoshihiro Saegusa

25.03.2019 Campus Talk in HKUST, CUHK
26.03.2019 Campus Talk in HKU
Presented by Yoshihiro Saegusa

A3966f193f4bef226a0d3e3c1f728d7f?s=128

LINE Developers
PRO

March 25, 2019
Tweet

More Decks by LINE Developers

Other Decks in Technology

Transcript

  1. Infrastructure in LINE LINE Corporation Yoshihiro Saegusa

  2. • Now responsible for infrastructure platform (private cloud) • Joined

    NHN Japan in April 2005 as a network engineer • Was responsible for networks and data centers until June 2017 ABOUT ME Yoshihiro Saegusa
  3. LINE Service Architecture Redis HBase Talkserver Object 
 Storage Image,

    File, Video VoIP BOT East Asia Users Southeast Asia Users North America Users Gateway(Legy) JP SG US TCP TCP TCP Gateway(Legy) Gateway(Legy)
  4. Physical Infrastructure Internet Datacenter Datacenter Physical
 Machine Datacenter Interconnect Rack

    Load Balancer Router Switch Datacenter
 Network
  5. Infrastructure Overview Physical Infrastructure Infrastructure Platform Application Application Application Application

    Application Application Application Application API CLI WebUI Infrastructure Private Cloud
  6. LINE Infrastructure Scale • 1Tbps+ User Traffic • 40,000+ Physical

    Machines • Multiple datacenters
 in 5+ countries
  7. CHALLENGES ɾɾɾ • 3,200 physical machines (10Gbps NIC) • 16,000

    Gbps of capacity • 2N redundancy Network POD Server ToR Distribution switch POD Scale • Network bottleneck • Inefficient architecture
  8. SOLUTIONS No bottleneck with CLOS network architecture ɾɾɾ ɾɾɾ Non-blocking

    large-scale network POD 7,200 physical machines Server Top of Rack (ToR) Leaf Spine 72,000Gbps capacity
  9. CLOS NETWORK EXAMPLE 2 Network PODs with 948 Switches 180

    server racks (7,200 servers)         180 server racks (7,200 servers)
  10. • applications running on private cloud • scale is a

    root of problems • keep improving physical infrastructure SUMMARY