Recommendations
Use QFX5130-32CD, QFX5140-24CD8O, or QFX5240-64OD switches as frontend leaf nodes.
Use QFX5220-32CD, QFX5230-64CD or QFX5240-64OD switches as frontend spine nodes.
Use HPE Juniper Apstra Data Center Director to automate deployment and monitor the frontend fabric.
Implement the solution following the minimum recommended software releases:
- All QFX devices: Junos OS Evolved 25.2X100-D20.4-EVO
- HPE Juniper Apstra Data Center Director: 6.1
To replicate the benchmarking testing described in this document use:
SGLang 0.4.5
SGLang Router 0.1.4
Envoy Proxy Service 1.35.3
NVIDIA GenAI-Perf 0.0.11
Llama 3.1 8B, Llama 3.3 70B, and Qwen 2.5 72B