Nebius Networking, Interconnects & Storage interview questions

Derive a complete rail topology and distinguish physical links, switch ports and transceivers. Size storage and management from their own requirements.

Two tenants can fit the average storage budget and still miss both checkpoint deadlines. Work through reservations, borrowing and bounded admission while protecting training reads, then test the design under failure.

Choose between supported InfiniBand and RoCE designs using workload tests, congestion behavior and a complete cost model.

Separate the all-reduce result from NCCL execution choices, and compare latency models without treating them as universal selection rules.

Track actual values across ranks, then count bytes with an explicit buffer definition. Reduction combines values; gathering and all-to-all change ownership.

Count endpoint, leaf, aggregation and core ports for an explicit 4,096-GPU design. Compare 1:1 and 2:1 without confusing switch ports, links or optics.

Design a durable dataset and optional working-set cache around access semantics, request overhead and versioned readiness.

Trace registered-memory access and GPU staging, then compare RDMA with sockets using measured CPU and application performance.

A bandwidth regression narrows the search only after the experiment is defined. Work through rank-count normalization, controlled path tests and the difference between collective time and training progress.

Calculate queue growth during incast, then distinguish endpoint rate feedback from link-level priority pause and shared-resource contention.

Derive bandwidth from actual per-rank traffic and readiness deadlines, showing why bucketing cannot erase a capacity shortfall.

Follow corresponding GPU positions through a rail, including the leaf and spine links needed when one rail outgrows a switch.

Compare serialized and overlapped storage paths, then test whether direct GPU I/O improves the actual pipeline stage.

Design snapshot ownership, background writes and manifest publication separately, with explicit checkpoint bytes and recovery scope.

Compare tensor, pipeline and expert traffic with per-request KV transfer. Work through declared payload and latency assumptions before choosing a serving network.

Measure batch readiness and producer capacity, then size prefetch for bursts without hiding a sustained throughput deficit.

Plan a reproducible tokenization build: measure CPU demand, choose a safe token dtype, publish complete shards and distinguish token lookup from training resume.