Network Topology
High-bandwidth, low-latency connectivity for distributed AI training and inference.
Network Architecture
Multi-layer network design optimized for GPU workloads with maximum throughput and minimal latency.
Layer 1 — Internet Edge
DDoS Scrubber
WAF
API Gateway
CDN
Layer 2 — Spine Layer
Arista 7800R3 Spine x2 (400 Gbps)
Layer 3 — Leaf Layer
Arista 7060X Leaf Switches (100/200 Gbps)
Layer 4 — GPU Compute Nodes
ConnectX-7 NICs
InfiniBand NDR (200G/400G)
Layer 5 — Storage
NVMe-oF
S3 Gateway (100 Gbps)
Architecture Details
In-depth specifications for each network layer.
External / Edge Layer
Dual ISP peering, BGP anycast, DDoS protection, SSL termination for secure external connectivity.
Core / Aggregation
Spine-leaf 400G architecture, VXLAN overlays, SDN controller for dynamic traffic management.
Compute / Access
Dual 200G ConnectX-7, InfiniBand NDR 400G, RoCE v2 for high-performance GPU communication.
Storage Layer
All-NVMe distributed storage, S3-compatible object storage, Network Volumes for persistent data.
Bandwidth Specifications
Detailed bandwidth allocation across all network layers.
| Layer | Components | Bandwidth |
|---|---|---|
| Internet Edge | DDoS Scrubber, WAF, API Gateway | 2 x 100 Gbps (BGP) |
| Spine Layer | 2 x Arista 7800R3 | 400 Gbps per spine |
| Leaf Layer | N x Arista 7060X | 100/200 Gbps per leaf |
| GPU Nodes | Dual ConnectX-7 NICs | 200G Ethernet / 400G IB |
| Storage | NVMe-oF arrays | 100 Gbps NVMe-oF |
| Management | OOB switches, IPMI | 1 Gbps dedicated |