Networking @Scale Fall 2022
as the organisers put it
Networking @Scale is a technical conference for engineers that build and manage large-scale networks. Networking solutions are critical for building applications and services that serve billions of people around the world. Building and operating such large-scale networks often present complex engineering challenges to solve. The Networking @Scale community focuses on bringing people together to discuss […]
who it's for
data center network engineers, wide area backbone engineers, network operators, optical network engineers, switching software engineers, network reliability engineers, site reliability engineers
topics
RDMA for large AI clusters; single-chip whitebox platforms; SAI-based switch software; MEMS optical circuit switching; aluminum subsea cable design; SRv6 wide-area routing; eBPF-based traffic sampling (Millisampler); network SLOs with synthetic blocks; OneWAN SDN traffic engineering; L4 routing consistency
| title | author | length |
|---|---|---|
| Opening Remarks | Rajiv Krishnamurthy | 07m |
Conference opens by mapping three concrete themes: data center, WAN, operations. Data center discussion highlights two decade shifts — tens of megawatts per… | ||
| Predictable Network | Dennis Cai | 27m |
Alibaba Predictable Network uses a Host-Network Fusion design to deliver predictable performance for large AI clusters, running large-scale RDMA for five years… | ||
| Developing, Deploying, Operating In-House Network Switches at a Massive Scale | Shrikrishna Khare, Srikrishna Gopu | 19m |
FBOSS is Meta’s in-house switch software rewritten to a single SAI-based implementation that lets Meta onboard a second ASIC vendor and run one unified binary… | ||
| Transforming Google's Datacenter Network Through Optical Circuit Switching | Rui Wang | 19m |
Google's Jupiter delivers a direct-connect mesh using MEMS-based Optical Circuit Switches as an indirection layer and a centralized Orion SDN control plane to… | ||
| Q&A Session 1 – Luca Niccolini | Luca Niccolini | 21m |
Cloud operators describe a predictable network built by co-designing software and hardware: a uniform single-chip whitebox platform running a common control… | ||
| Wiring the Planet: Scaling Meta's Global Optical Network | Stephen Grubb, Joseph Kakande | 24m |
Meta builds and operates an expanding global optical mesh: more than 150,000 route kilometers of subsea cable and dozens of projects with 3,600+ route… | ||
| OneWAN Traffic Engineering: Unifying Inter-Datacenter and Internet Backbones | Umesh Krishnaswamy | 14m |
Microsoft consolidates AS8075 and AS8074 into OneWAN, a SDN-controlled backbone that connects 60+ Azure regions and 190 PoPs and handles 10× the devices and… | ||
| Q&A Session 2 – Andrey Golovanov | Andrey Golovanov | 23m |
Aluminum subsea cable math: aluminum is 61% as conductive as copper, 30% of the weight and 25% of the price, so cables use more aluminum to match resistivity… | ||
| Millisampler: Fine-Grained Network Traffic Analysis | Yimeng Zhao | 21m |
Millisampler is a lightweight eBPF-based profiler that samples per-CPU traffic at 100µs–10ms granularity by running a TC filter in kernel space and a… | ||
| Network SLOs: Knowing When The Network Is The Barrier To Application Quality Of Experience | Brandon Schlinker, Sharad Jaiswal | 23m |
Meta defines concrete Network SLOs by measuring passive transfer times of synthetic blocks (not speedtests) and choosing the best predictor via AUC to map… | ||
| Responsiveness Under Working Conditions | Christoph Paasch | 24m |
Defines Responsiveness under working conditions as HTTP/2/3 latency measured while saturating the link with concurrent HTTP bulk transfers, sending HTTP GET of… | ||
| Q&A Session 1 – Neil Spring | Neil Spring | 21m |
Q&A covers concrete results and operational tools: Millisampler shows short, frequent microbursts that cause loss when contention is high, with contention… | ||
| Improving L4 Routing Consistency | Aman Sharma, Andrii Vasylevskyi | 21m |
Layer 4 load balancing misroutes when both a load balancer and a backend change because consistent hashing plus per-host LRU caches lose connection affinity… | ||
| Highly-Available, Global Traffic Managmement at Scale | Alan Halachmi, Akshat Aranya | 16m |
AWS Global Accelerator presents two static IPv4 addresses (optionally two IPv6) advertised via anycast from 104 PoPs, terminates client TCP at the edge to… | ||
| Q&A Session 2 – Bharat Parekh | Bharat Parekh | 22m |
RTMP third‑party encoders stay on TCP while first‑party mobile streams use Rush over QUIC; standardization via the IETF Media over QUIC working group is… | ||
| nothing matches | ||