Networking @Scale 2023
2023-09-07
-- online --
[Networking @Scale 2015, Networking @Scale Summer 2022, Networking @Scale Fall 2022, Networking @Scale 2023, Networking @Scale 2024]
as the organisers put it
Join us for Networking @Scale 2023 for engineers managing large-scale networks. Solve complex challenges in building and operating networking solutions.
who it's for
data center network engineers, network architects, AI infrastructure engineers, HPC engineers, site reliability engineers, network operators, network performance engineers
topics
RDMA backend network design; RoCEv2 with PFC and ECN; traffic engineering controllers; leaf-spine CLOS fabric design; large GPU cluster networking; network observability for training; AI system performance simulation; NIC GPU CPU host tuning; flow multiplexing and QP hashing; DCQCN congestion control; GPUDirect peer memory; 100G to 400G link upgrades
| title | author | length |
|---|---|---|
| Welcome Remarks | Rajiv Krishnamurthy | 07m |
Announces a dedicated backend network (RDMA) built separate from front-end services, designed to support tens of thousands of GPUs, exabytes of storage and… | ||
| Networking for GenAI Training and Inference Clusters | Jongsoo Park, Petr Lapukhov | 23m |
Large language model workloads drive orders of magnitude more compute and new network designs: training examples cite a 70B-parameter Llama2 trained on ~2… | ||
| Meta’s Network Journey to Enable AI | Hany Morsy, Susana Contrera | 17m |
Describes concrete network choices: moves from a 144‑GPU single‑BTSW star (single‑point‑of‑failure) to a two‑stage CLOS with RTSW/CTSW rack and cluster… | ||
| Traffic Engineering for AI Training Networks | Shuqiang Zhang, Jingyi Yang | 17m |
Meta runs RoCE training clusters with up to 4K GPUs in a two-layer “factory” topology (RTW/CTSW) where each RTW has 16 downlinks; static DPR routing caused big… | ||
| Live Q&A with Speakers – James Zeng | James Zeng | 23m |
Live Q&A gives concrete operational answers: 7:1 (16:1 feasible) super‑spine over‑subscription because most traffic stays inside pods or racks; 400G copper DAC… | ||
| Scaling RoCE Networks for AI Training | Adi Gangidi | 21m |
Meta runs RoCEV2 on commodity Ethernet to support AI training fabrics with tens of thousands of GPUs, using servers with 8 GPUs and 8 NICs and 16 uplinks per… | ||
| Network Observability for AI/HPC Training Workflows | Shengbao Zheng | 18m |
ROCET associates jobs to RDMA hardware counters (out‑of‑sequence errors, NIC flap, local ACK timeouts) and correlates those counters with training throughput… | ||
| Arcadia: End-to-end AI System Performance Simulator | Zhaodong Wang, Satyajeet Singh Ahuja | 17m |
Arcadia simulates compute, memory and network together to predict AI cluster behavior and outputs training and inference metrics, utilization and reliability… | ||
| Live Q&A with Speakers – Joseph Provine | Joseph Provine | 15m |
Live Q&A reveals concrete systems and numbers: a deployment using 16×200G NICs injected into 8×400G uplinks to reduce ECMP hash collisions; GPUDirect peer… | ||
| nothing matches | ||