The bridge
TCP slows down when the network is congested. A microservice should do the same when downstream is saturated — but we often keep accepting requests until everything times out.
BGP picks routes based on policy and reachability. Service meshes and API gateways do something similar: route around unhealthy instances, prefer local paths, fail over when a hop is down.
The vocabulary transfers. The implementation is different. That's the point — networks coursework is not a detour from backend engineering.
What I saw in production
At Cognizant, a hot path query ran at P99 ~250ms under load. Indexing and query shape fixed it to ~50ms — but the symptom looked like congestion: requests piling up, threads waiting, timeouts at the edge.
AlgoChronicle's GitHub Actions pipeline is a different layer, but the same instinct: don't let unbounded work enter the system. Parse, validate, sync, then update the UI — each stage has a boundary.
JobHuntOS LLM routing is failover routing. When Claude throttles, walk the chain — same mental model as routing around a down BGP peer, except the 'peer' charges per token.
What I'm exploring next
One infra artifact: Terraform module or observability write-up for Neocortex deploy — showing how I think about deploy and ops, not just app code.
Publish this note on LinkedIn and link it from the portfolio writing section.
Keep connecting networks theory to backend practice in system design conversations.