AI: Network
In one line Precise time sync (PTP / IEEE 1588) gives the fabric a common, sub-microsecond clock - essential for correlating telemetry, measuring congestion, and coordinating distributed workloads. PTP basics Hardware timestamping, a grandmaster clock, and boundary / transparent clocks in the path deliver sub-microsecond accuracy - versus NTP's millisecond
AI: Network
Requirement / business driver Decide how the AI fabric is structured and operated - a dedicated island vs part of the DC, traditional routing vs SDN, single vs multi-site. Connectivity models (2.8) * Isolated / purpose-built island - a dedicated back-end fabric for GPU/RDMA traffic, separate from front-end, management,
AI: Network
A concrete, tested anchor for the abstract fabric notes: Cisco's Validated Design for a non-blocking, low-latency, lossless Ethernet fabric carrying RoCEv2 for AI/ML. It pins the topology math, the eBGP underlay, the server-attach model, and the QoS recipe (Lossless fabric - PFC, ECN, DCQCN) into
AI: Network
The bridge between the workload and the fabric: how a model is split across GPUs decides the traffic the network must carry. A network architect doesn't train the model, but must know which parallelism strategy and which collective operations produce the east-west load (AI-ML workloads and
AI: Network
AI backend traffic breaks the assumption enterprise ECMP relies on. Enterprise links carry many small flows that hash evenly; an AI backend carries a handful of massive elephant flows (GPU-to-GPU collectives at line rate, often Tb/s). Spreading those few flows well is a first-class design problem.
AI: Network
A training job spreads across hundreds of GPUs, so the fabric must keep running through device failures. The good news of scale: *the bigger the fabric, the higher the chance something fails - but the smaller each failure's blast radius.* Spine failure - graded, not catastrophic In a non-blocking
AI: Security
Requirement / business driver The hosting model sets the entire security posture of an AI environment - choose it against sensitivity, sovereignty, and threat exposure, not just cost. Options * Air-gapped - totally isolated, no external connectivity; for classified / sovereign / highly sensitive training. Highest security, lowest agility. * Private (on-prem / private cloud) - enterprise
AI: Security
In one line You cannot drop a firewall into the middle of a lossless RDMA fabric - so AI back-end security comes from isolation and segmentation, not inline inspection. The tension RoCE/RDMA is line-rate, latency-sensitive, and lossless (RDMA, RoCE and RoCEv2, Lossless fabric - PFC, ECN, DCQCN). Inline
AI: Security
Requirement / business driver Safely share extremely expensive GPU infrastructure across tenants/teams without lateral movement or data leakage between them. Why it's hard GPUs are too costly to dedicate per tenant, so clusters are shared - which means each tenant's compute, RDMA traffic, and storage must be
AI: Security
In one line The model and the training data are the crown jewels - protect their confidentiality and integrity across the whole lifecycle, not just the network around them. The threats * Model theft / exfiltration - the trained model is valuable IP; protect against extraction. * Training-data confidentiality - datasets often contain sensitive/regulated
AI: Security
Requirement / business driver Inference endpoints and APIs are user/internet-facing services - protect them against malicious use, the way you would any exposed application, plus AI-specific abuse. The surface Inference is served via APIs/apps (often internet-facing, north-south, AI-ML workloads and their infrastructure impact). This is
AI: Security
Note: Adjacent scope This is AI used as a defensive tool, distinct from securing the AI fabric (Securing the high-performance fabric). It maps to the blueprint's "protection against malicious uses" and to Core zero-trust policy-building with AI/ML - a common exam-adjacent AI