Network Design & High Availability

Each leaf loses one of six uplinks: 600 Gb/s becomes 500 Gb/s, about a sixth of fabric capacity, and a 2:1 leaf becomes 2.4:1. No server loses reachability; width, routed links and ECMP keep the failure small.

Convergence time is the sum of failure detection, propagation to other routers, path computation and forwarding-table install. Give each phase a measured or assumed value, add them with a margin, and compare against the one-second limit.

After one link fails, the survivor must carry both loads, 2 × 65% = 130% of its capacity, so it congests and drops traffic. Keep each member of a pair below about 50% at peak and order upgrades ahead of lead times.

Spanning tree leaves each access switch forwarding toward the VLAN's root. If VRRP's Active Router is the other distribution switch, every upstream frame crosses the inter-distribution link; co-locating root and Active keeps the path one hop.

VRRP sees only LAN advertisements, so a router with a dead uplink stays Active. Tracking, an implementation feature, cuts its priority below the Backup's; with preemption on, the Backup takes over. A flapping uplink then moves the role back and forth.

Once detected, the member leaves the bundle, its in-flight frames are lost and its flows rehash onto the survivors, leaving 30G. A minimum-links threshold takes the whole bundle down below N members so routing moves traffic to a healthier path.

Two switches present one shared LACP System ID, so the server bundles links to both as to one partner. If their peer link fails, a separate keepalive shows the peer is alive and the secondary disables its MLAG ports, avoiding split-brain.

Tail drop discards only when a queue is full, so many flows lose packets at once, slow down together and leave the link idle. RED drops randomly and early as the average queue grows; weighted RED sets thresholds per drop precedence.

Local breakout sends chosen applications straight from each store to the internet, cutting the detour, data-centre bandwidth and a shared choke point; inspection, logging, address translation and policy that lived at the data centre must now exist per store.

Copy only the shared-services prefix into each zone VRF and only each zone's own prefix back into the shared VRF, filtered so nothing passes onward. Overlapping zone prefixes need translation or renumbering, and leaked paths bypass any firewall.

Route at every leaf with ECMP across the spines, adding a VXLAN overlay with BGP EVPN only where workloads need Layer 2 reach. Stretched Layer 2 eases VM mobility but makes the broadcast domain the failure domain.

Two independent 99.9% circuits do reach 99.9999% as a pair, but the site is the whole path: a single edge router, power feed or shared duct in series caps it far lower. Find those elements, quantify each, fix the largest.

MPLS L3VPN wins where predictability is the requirement: one provider engineers the whole path, honours traffic classes under a contract and gives any-to-any reachability; internet tunnels win on price, speed of delivery, provider diversity and direct cloud access.

Find the congestion points, mark at the trust boundary, shape to the contracted rate, give voice a policed priority queue sized from calls times per-call rate, give other classes minimum shares, and cap calls with admission control so the policer never drops.

Tier stores by outage cost and give each physically diverse transports, such as dual broadband plus LTE, or MPLS plus internet at critical sites; you give up a contracted end-to-end bound on loss and latency for measurement and steering.

Both can shrink PCI DSS scope. A payment VRF on shared routers is cheap but puts every router and switch carrying it into scope and relies on configuration discipline; separate hardware costs more but keeps scope and blast radius small.

BFD echo packets are addressed so the peer simply forwards them back, testing its forwarding path without its BFD process answering; echo needs the peer's consent, works only on single-hop sessions and breaks under ingress filtering.

A VRRPv3 Backup declares the Active down after 3 x interval + Skew_Time, about 3.61 s at the 1-second default for priority 100. Centisecond intervals cut that below 40 ms, but risk false failovers under queueing delay and complicate mixed VRRPv2 operation.

RSVP-TE fast reroute pre-signals a backup path around each protected link or node; when the failure is detected, the adjacent router redirects traffic onto that backup immediately, with no path computation or signalling, while the head-end later re-optimises.