Senior Network Production Engineer, AI Supercomputing

Microsoft AI•Redmond, WA

10d•$119,800 - $304,200

About The Position

We are building and operating frontier-scale AI supercomputers used to train the world’s most advanced models. The backend network is a critical part of the machine: a single degraded NIC, switch, link, or path can reduce training performance or disrupt jobs spanning thousands of GPUs. We are looking for a senior, hands-on Network Production Engineer to own the production health and operations of our multi-rail Ethernet backend network built on MRC and RDMA technologies. You will bridge network engineering, distributed systems, hardware health, and production operations to ensure the network delivers predictable performance at extreme scale. Your mission is simple: make the network invisible to researchers by maximizing large-job success, minimizing performance degradation, and recovering quickly when failures occur.

Requirements

Bachelor’s Degree in Computer Science, or related technical discipline AND 4+ years technical experience in Site Reliability Engineering, DevOps, or Infrastructure Engineering OR equivalent experience
Experience operating large-scale datacenter or high-performance computing networks in production environments.
Experience with Ethernet, RDMA, RoCEv2, congestion control, routing, load balancing, and lossless or near-lossless network design.
Experience diagnosing failures across switches, NICs, hosts, drivers, firmware, and distributed applications.
Experience working with Linux systems and programming in Python, Go, C++, or another systems-oriented programming language.
Experience building monitoring, automation, diagnostic, or remediation systems for production infrastructure.
Experience leading high-severity incident response and coordinating resolution across multiple engineering teams.
Experience analyzing telemetry to identify and validate root causes of infrastructure or network issues.
Ability to participate in a global on-call rotation.

Nice To Haves

Master’s Degree in Computer Science, or related technical discipline AND 2+ years technical experience in Site Reliability Engineering, DevOps, or Infrastructure Engineering OR equivalent experience
Experience operating GPU training clusters or other tightly coupled distributed-computing systems.
Knowledge of MRC or other multi-rail, multi-plane, multipath Ethernet transports.
Experience with NCCL, collective communications, CUDA, GPUDirect RDMA, and distributed training frameworks.
Familiarity with ConnectX-class NICs, modern Ethernet switch ASICs, SONiC, SAI, and switch telemetry.
Experience with InfiniBand and the operational differences between InfiniBand and Ethernet-based AI fabrics.
Familiarity with Kubernetes, Slurm, workload placement, and cluster-health certification.
Experience with streaming telemetry, gNMI, Prometheus, Datadog, Kusto, eBPF, or equivalent observability systems.
Experience managing fleet-wide firmware, driver, or network operating-system changes.
A track record of improving reliability through automation rather than recurring manual intervention.

Responsibilities

Own the availability, performance, and operational readiness of the MRC Ethernet backend network.
Define and operate service-level indicators for packet loss, congestion, link health, path diversity, bandwidth, tail latency, collective performance, and job impact.
Detect slow or degraded components before they cause training failures or reduce model FLOPs utilization.
Build health models that correlate switch, NIC, host, topology, and application telemetry.
Participate in on-call rotation and lead the response to high-severity network incidents.
Diagnose failures spanning GPUs, NICs, switches, cables, firmware, drivers, network operating systems, MRC, NCCL, Kubernetes, and training workloads.
Develop automated mitigation mechanisms, including path avoidance, node quarantine, workload relocation, and safe component remediation.
Create operational procedures for maintenance, upgrades, rollback, capacity expansion, and topology changes.
Work directly with training teams to investigate collective-performance degradation, stalls, timeouts, job restarts, and unexplained MFU loss.
Translate low-level network signals into clear job-level impact.
Establish network qualification gates for admitting nodes and racks into large training pools.
Run failure-injection and resilience testing to validate behavior under link, NIC, switch, and path failures.
Improve placement and routing policies for jobs spanning racks, rows, and network planes.
Build production-quality software and automation for network validation, monitoring, diagnosis, remediation, and fleet-wide change management.
Replace manual investigation with deterministic workflows and actionable alerts.
Automate firmware, driver, configuration, and network operating-system compliance checks.
Maintain authoritative network inventory, topology, and configuration state.
Reduce mean time to detect, isolate, mitigate, and permanently resolve failures.
Lead technical investigations involving internal infrastructure teams, cloud and datacenter operators, Azure Networking, NVIDIA, switch and NIC teams, and network-software partners.
Own incidents through resolution, even when the underlying failure crosses organizational boundaries.
Produce clear root-cause analyses with corrective actions, owners, and completion dates.
Convert recurring production failures into requirements for future network, server, and accelerator architectures.
Create runbooks and train engineers across regions to provide reliable round-the-clock coverage.