Design, implement, and operate high-performance network fabrics for large-scale GPU clusters and AI training workloads, using InfiniBand and high-speed Ethernet (200/400GbE).
Responsibilities:
- Design and deploy leaf–spine topologies; configure switches and build RoCEv2/InfiniBand fabrics.
- Implement network segmentation and multi-tenancy; integrate DPUs, SR-IOV vNICs, and networking offloads.
- Monitor fabric health, troubleshoot AI training pipelines, manage routing and security configurations, and maintain switch upgrades and firmware.
- Automate network configuration and contribute to tooling, dashboards, alerting, fabric validation, auto-provisioning, and drift detection.
- Deploy and optimize NVMe-over-Fabrics and high-performance storage networking; collaborate on data center rack integration and testing.
- Implement secure multi-tenant isolation and harden APIs, management networks, and out-of-band access.
Requirements:
- Deep expertise in InfiniBand HDR/NDR or high-speed Ethernet (200G/400G), and strong RDMA knowledge including RoCEv2 and congestion management.
- Advanced routing and switching experience with BGP, EVPN, OSPF, and MLAG/VPC; experience with Mellanox/NVIDIA platforms.
- Strong Linux networking fundamentals and experience supporting AI training clusters using H100, B200, A100, or similar GPUs.
- Familiarity with BlueField DPUs, SR-IOV, VirtIO, and OVN/Kube-OVN/Calico networking.
- Understanding of NVMe-oF over TCP or RDMA. Experience with WekaFS, DDN, VAST, or Pure is an advantage.
- Kubernetes networking, OpenStack Neutron/OVN, and GPU virtualization or MIG networking experience are nice to have.