Keynote: Enterprise Data Center Migration to Hyperscale – Zaid Ali Kahn
summary
LinkedIn describes a concrete migration: a custom Altair rack (96 hosts) design with single‑SKU 1U whitebox switches, four independent color fabrics and 4‑way ECMP that drops per‑packet chip traversal from ~15–20 to <5 chipsets per packet. Fabric supports 100,000+ bare‑metal servers, targets <6:1 oversubscription and host speeds of 10/25/50/100 Gb. Spine uses 100G split into four 50G paths via PSM4. Telemetry streams into a Kafka pipeline for machine‑learning driven self‑healing and streaming telemetry via Kafka automation.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:46Traffic Demands
- 1:59What is invisible software infrastructure
- 3:04Linkedin Platform as a Service LPS
- 4:13Key infrastructure visions
- 5:27Altair flagship design
- 6:37Scaling out characteristics
- 7:53Kill chassis based devices
- 8:31Introducing parallel fabrics
- 10:09Datacenter fabric architecture
- 10:31100G transformation
- 11:36Network Operating System
- 12:57OpenFabric Control Plane
- 14:21Rethinking The Network Stack
- 14:51Management Plane: Reducing Protocols
- 15:24Programmable datacenter
related talks
17mMeta’s Network Journey to Enable AI
21mDesigning Scalable Networks for Large AI Clusters: Challenges and Key Insights
15mAlibaba HPN: A Data Center Network for Large Language Model Training
26mApplying New High-Throughput Network Solutions in IEC (Integrated Edge Cloud)
15mKeynote: Scaling DevOps – Moving from Mechanization to Automation
16mMatryoshka: Realizing Hyperscale Data Center Network Design for the AI Era
44mHow we designed an OpenStack Public Cloud at Scale with Fabric Networking
14mA Journey or There and Back Again