Broadcom Jericho3 AI Ethernet Fabric

Sdílet
Vložit
  • čas přidán 30. 07. 2023
  • Broadcom offers scheduled fabric solutions, with a focus on switch-based architecture using the Jericho and Ramon families. This architecture ensures congestion-free operations and perfect load balancing for AI workloads. The leaf devices handle switch functionality, lookup, queuing, and scheduling, while the spine devices interconnect the leaf devices and are optimized for AI workloads. The fabric allows for traffic segmentation and reassembly in cells, with hardware-based mechanisms ensuring no impact from network failures. The scalable architecture can support networks ranging from a few hundred boxes to tens of thousands, with the latest generation devices allowing for scaling from 18K to 32K nodes and 400GB to 800GB Ethernet. It provides multi-tenancy and job isolation for AI clusters, and offers hardware-based telemetry and diagnostic tools for visibility into network performance. The future-proof design ensures adaptability to evolving AI models and workloads.
    Presented by Henry (Xiguang) Wu, Product Line Manager. Recorded at Networking Field Day 32 on July 26, 2023 in San Francisco, CA. Watch the entire presentation at techfieldday.com/appearance/b... or visit techfieldday.com/event/nfd32/ or www.Broadcom.com/ for more information.
  • Věda a technologie

Komentáře • 5

  • @aaron-az
    @aaron-az Před 10 měsíci +1

    Reminds me of ATM LAN emulation from back in the '90's. VBR anyone? Yes I'm old. 😉

    • @Arvizio
      @Arvizio Před 8 měsíci

      Ha yes, I'm old too.....

  • @louissenderler6866
    @louissenderler6866 Před 9 měsíci +3

    Just like Cisco CELL (64byte fixed length) transmiting over GSR FABRIC, Cisco put the scheduler in a LC... this is now the popullar way building DC fabric ( is Google TIN same?).

  • @nomayor1
    @nomayor1 Před 3 měsíci

    It appears to me as if Broadcom took a Juniper router from thirty years ago and distributed its line cards and switch fabric cards into leaves and spines.

  • @dubkayc
    @dubkayc Před 9 měsíci

    How do they achieve 10ns convergence times esp when there may be indirect link failures, say, between the egress leaf to spine link?
    The cells in flight between the ingress leaf to spine will be blackholed as well as any traffic until the ingress leaf realizes the topology change and sends all traffic to the other spine exclusively.