Senior RDMA Performance Engineer

Role overview

Validate and tune data center networks built for AI workloads, with a focus on RDMA performance on Spectrum-based platforms. The work combines switch and NIC configuration, congestion control, benchmarking, and analysis of network telemetry to improve throughput, latency, and job completion time. The project follows Agile practices in a startup-style environment with a US-based client.

Responsibilities - Configure, troubleshoot, and maintain Spectrum switches and ConnectX network interface cards. - Implement and tune RoCEv2, Priority-based Flow Control, and Explicit Congestion Notification for AI traffic. - Optimize network setup, including switch buffers and quality-of-service policies, for throughput, latency, and job completion time. - Benchmark and validate AI data center fabrics using RDMA and the NVIDIA Collective Communication Library (NCCL). - Use AI agents and specialist tools to automate routine network work, testing, and telemetry analysis. - Document network designs, configurations, and benchmark findings.

Requirements - Seven or more years of network engineering experience; three or more years in data center design and architecture is preferred. - Hands-on experience with BGP, EVPN/VXLAN, LACP, and ECMP. - Deep understanding of RoCEv2, PFC, and ECN in AI data center environments. - Experience tuning switch buffers and QoS policies. - Background benchmarking and validating RDMA fabrics with NCCL. - Ability to use AI tools for automation and analysis, produce clear technical documentation, and communicate professionally in English.

Nice to have

Advanced networking certifications, network automation and validation tools such as pyATS, NAPALM, Batfish, Nornir, or NetBox, and experience with CI/CD or infrastructure as code.

Benefits and work setup

Flexible hours and remote, office, or hybrid work are available, along with hands-on onboarding, internal training, a training budget, and the possibility of changing projects.