Own the switch software and firmware lifecycle—Cumulus Linux, SONiC and switch-OS upgrades, CPLD and field-notice rollout campaigns—with staged, tested upgrade plans that keep partner fabrics off stale releases without disrupting production workloads.
Support Day 1 fabric validation and acceptance: InfiniBand and UFM bring-up, Spectrum-X/RoCE Ethernet configuration, cabling and link-health verification, routing and congestion-control validation, and multi-day, multi-rack burn-in against agreed MTBI and goodput targets.
Minimise the time from cluster handover to first production workload, removing duplicated fabric validation across hardware bring-up, managed-service intake and the partner’s own network operations teams.
Drive fabric reliability at fleet scale: telemetry for link flaps, retransmits and congestion, fault detection through to remediation, root-cause analysis of network-induced job failures, and measurable improvement of MTBI and job goodput.
Provide consultative guidance and hands-on troubleshooting across the fabric stack—NICs and DPUs, switch OS, routing and congestion control, host networking and Kubernetes network integration—and act as the technical leader for assigned accounts, running structured knowledge transfer and producing runbooks so partner teams can operate advanced fabric configurations independently.
What We Need to See
BS/MS/PhD in Computer Science, Electrical/Computer Engineering, Physics, Mathematics, or related fields, or equivalent experience.
5+ years of professional experience in data centre networking, fabric engineering or large-scale network operations roles.
Fabric & Data Centre Networking: Deep understanding of data centre network architectures and RDMA fabrics—InfiniBand and RoCE/Ethernet—including topology and routing design (fat-tree, rail-optimised), congestion control, lossless/QoS configuration, and troubleshooting across NICs, switches and high-speed interconnects.
NVIDIA Fabric Technologies: Hands-on experience with InfiniBand and UFM, Spectrum-X Ethernet, Cumulus Linux and/or SONiC, ConnectX and BlueField NICs/DPUs, and NVLink/NVSwitch systems—ideally including NMX-C / NMX-M partition operations on NVL72-class platforms.
Linux & Switch Platforms: Deep knowledge of Linux (RedHat, Ubuntu) and switch operating systems, network OS internals, host networking and OS-level security, together with an understanding of the storage and compute traffic patterns of HPC/AI clusters.
Automation, GitOps & Observability: Proficiency in Python and Bash scripting, configuration management and Infrastructure-as-Code tools (e.g. Ansible, Terraform), GitOps-based network configuration and firmware/upgrade management for large fleets, and observability stacks (Grafana, Loki, Prometheus) applied to fabric telemetry.
Fleet Reliability & Customer Engagement: Demonstrated ability to measure and improve MTBI and job goodput on large GPU clusters—fault detection, drain and remediation workflows, SLO/error-budget definition and post-incident review—combined with a strong consultative background leading architectural reviews and presenting to executive stakeholders.
Ways to Stand Out from the Crowd
Experience operating Kubernetes networking in GPU clusters—CNI plugins, multi-network attachment, SR-IOV and RDMA device plugins—and with the NVIDIA Network and GPU Operators for automated resource lifecycle management.
Expertise with DOCA and DPU infrastructure services (e.g. DOCA DPF) for offloaded networking, security and storage services.
Familiarity with fabric and GPU health telemetry—DCGM and XID diagnostics, link-level and switch counters, node-level health agents, and fleet-wide reliability intelligence.
Knowledge of large-scale training traffic behaviour: collective communication (NCCL) tuning, rail alignment, and diagnosing network-bound performance regressions in distributed training jobs.
Experience delivering multi-tenant network isolation for GPU-as-a-service estates—VRF/VLAN/EVPN segmentation, tenant partitioning and secure fabric handover between tenants.
