Overview
In this role you will shape the network architecture for k0rdent AI, Mirantis’ platform for GPU clouds and AI factories. You’ll design scalable data-center fabrics and multi-tenant isolation, then produce reference architectures, proofs of concept, and runbooks for real hardware with customers and partners. The role blends research with hands-on delivery, guiding both lab validation and customer-facing discussions. You will publish findings and influence the roadmap, working across global teams to push innovative AI networking designs.
Verantwoordelijkheden
- Design and publish network reference architectures for k0rdent AI across data center scales from single rack to thousands of GPUs
- Define compute fabrics (InfiniBand, RoCEv2 Ethernet) including rail-optimized and fat-tree/Clos topologies and related subsystems
- Specify multi-tenant isolation mechanisms (InfiniBand PKeys, VRFs, EVPN-VXLAN, Kubernetes network policy) and document trade-offs
- Harden host/platform networking: NIC exposure, DPUs, SR-IOV, IOMMU groups, NUMA, GPU-NIC affinity
- Design Kubernetes networking for AI: CNIs, Multus, SR-IOV, RDMA device plugins, NVIDIA Network Operator, Dynamic Resource Allocation
- Prototype and evaluate new designs in lab settings and publish results as internal or external papers/talks
- Build automation/tools to reproduce architectures (Python, Go, Bash, Ansible, Helm, Terraform) and run validation benchmarks
- Act as network SME in customer workshops, partner reviews, and joint design efforts with NVIDIA, OEMs, and vendors
- Contribute to product/engineering with research outcomes and help shape k0rdent AI roadmap
- Present at industry events and run hands-on technical workshops for diverse audiences
Belangrijkste vereisten
- 8+ years in network engineering or architecture with at least 3 years in data center, HPC, or cloud networking
- Customer-facing experience as solutions architect, presales engineer, consultant, or technical lead
- Data center network design experience (spine-leaf, Clos, rail-optimized GPU fabrics, oversubscription)
- InfiniBand subnet management, partitioning, adaptive routing, and NCCL/RDMA behavior
- RoCEv2 Ethernet with lossless designs (PFC, ECN, DCQCN), QoS, MTU, buffer tuning
- Routing/overlays (BGP, EVPN-VXLAN, VRFs) for multi-tenancy
- Hybrid/multi-cloud connectivity and IP/DNS planning across sites
- Linux networking basics (NICs, PCI passthrough, SR-IOV, IOMMU, NUMA) and tooling (iproute2, ethtool, devlink)
- Kubernetes networking (CNI plugins such as Cilium/Calico/OVN-Kubernetes, Multus, SR-IOV, RDMA devices, network policy)
- KubeVirt virtualization networking (passthrough, SR-IOV)
- Programming/scripting in Python or Go; Git, CI, and IaC familiarity
- Curious and data-driven mindset
- Ownership and accountability
- Proactive problem solving
- Data center network design: spine-leaf, Clos, rail-optimized fabrics, oversubscription, ECMP
- InfiniBand: subnet management, partitioning, adaptive routing, NCCL/RDMA behavior
- RoCEv2 on Ethernet: lossless designs, PFC/ECN, DCQCN, QoS
Senior Solutions Architect – AI Infrastructure Networking in Amsterdam werkgever: Jobrapido
StarPark is een uitstekende werkgever die je de kans biedt om te werken op diverse evenementen en festivals, waar je in een informele en prettige werksfeer kunt bijdragen aan de veiligheid van bezoekers. Met flexibele werkmogelijkheden en goede begeleiding vanuit het team, stimuleert StarPark persoonlijke groei en ontwikkeling, waardoor je niet alleen waardevolle ervaring opdoet, maar ook de mogelijkheid hebt om door te groeien binnen de organisatie. Wij verwelkomen iedereen, ongeacht achtergrond, en geloven dat jouw talent het belangrijkste is.
