AI: Compute, storage & environment
In one line AI infrastructure is built around accelerators (GPUs) kept busy by intelligent NICs (SmartNIC / DPU) that offload networking and storage so the expensive GPUs never wait. The components * GPU - the accelerator: massive parallel compute plus high-bandwidth memory (HBM). GPUs connect intra-server by NVLink/NVSwitch (scale-up)
AI: Compute, storage & environment
Requirement / business driver Connect many GPUs so collectives run fast - across two dimensions whose boundary shapes the whole fabric. The two dimensions * Scale-up (inside the node/domain) - NVLink / NVSwitch (and the emerging UALink - an open, vendor-neutral consortium standard; AMD's Infinity Fabric is the proprietary parallel) connect
AI: Compute, storage & environment
Requirement / business driver Feed training data and absorb periodic checkpoints without starving the GPUs or stalling on I/O. Two I/O patterns * Training-data ingestion - sustained, parallel high read throughput to keep GPUs fed. * Checkpointing - periodic, bursty, all-nodes-write-at-once (incast!) saves of model state for fault
AI: Compute, storage & environment
Requirement / business driver AI's power density and energy use are first-order design constraints - the facility's power and cooling envelope can cap how many GPUs you can deploy, and sustainability/cost pressures make efficiency a design driver, not an afterthought. Power density AI racks draw 40-
AI: Compute, storage & environment
In one line AIOps applies ML to the flood of telemetry and logs an AI fabric produces - to detect anomalies, correlate events, and (eventually) auto-remediate - because thousands of GPUs and sub-second telemetry are beyond human monitoring. The three blueprint pieces * Data scrubbing (4.3.a) - clean, normalize, and
AI: Compute, storage & environment
In one line Storage networks are designed around one question - how does the fabric guarantee delivery, and who pays for that guarantee - and each protocol answers it differently: FC builds losslessness into the link layer, FCoE borrows it from Ethernet's DCB toolkit, iSCSI lets TCP absorb loss, and
Core: Transport technologies
In one line The physical layer sets the distance, bandwidth, and resilience ceiling that every higher layer inherits - and cabling outlives the electronics plugged into it. Copper Twisted pair (Cat5e/6/6A), ~100 m channel limit, carries PoE / PoE+ / PoE++ for powering APs/phones/cameras, but is susceptible to EMI
Core: Transport technologies
In one line Ethernet is the universal L2 framing from the access edge to 400G+ and out into the WAN as Carrier / Metro Ethernet. Speeds, lanes, and optics 1 / 10 / 25 / 40 / 100 / 400G. Higher speeds use multiple lanes (40G = 4x10G, 100G = 4x25G, 400G = 8x50G PAM4), enabling breakout. At 25G
Core: Transport technologies
Requirement / business driver Carry capacity across distance - campus/metro links, data-centre interconnect (DCI), or fibre-exhaust relief - as cost-effectively as the growth curve allows. Constraints Number of wavelengths needed now and later, distance (amplification?), budget, own vs leased fibre, and operational capacity. Options * Grey optics over dark fibre
Core: Transport technologies
Requirement / business driver Choose the access / transport methods for sites and the WAN, balancing bandwidth, cost, availability/SLA, latency, and reach across very different site profiles. Options Leased line / private circuit, MPLS access or Metro Ethernet, broadband (xDSL / DOCSIS / FTTx), cellular (4G / 5G, incl. fixed-wireless and backup), satellite (high-
Core: Transport technologies
In one line Physical-layer protection and path diversity decide whether higher-layer convergence ever gets tested - and legacy transports must be migrated cleanly. Physical resiliency 1+1 / 1:1 protection, optical rings with protection switching, or mesh; dual entrance and diverse fibre paths. The trap is SRLG / fate-sharing:
Core: Layer 2 Control Plane
In one line A Layer 2 domain is a single broadcast and failure domain that floods to learn - so the master rule is keep it small and push routing toward the edge. How L2 behaves Switches flood-and-learn: unknown unicast/broadcast/multicast floods to every port in the VLAN,