Learning series · 6 episodes · Staff / Principal
The Network Was Up. The Requests Were Not.
Six networking incidents connecting TCP windows, packet loss, listen queues, NAT exhaustion, tunnel MTUs and gray failures.
Start series →Follow the decisions in order
- Part 1 of 6
Staff
The link is 10 Gbps. Why is one file moving at 200 Mbps?Calculate BDP and distinguish receiver-window, congestion-window and application pacing limits on a long-RTT TCP flow.
- Part 2 of 6
Staff
Only one percent of packets are lost. Why did API p99 explode?Diagnose how a small segment-loss rate can block ordered bytes, trigger retransmission timers and alter tail latency and congestion state.
- Part 3 of 6
Staff
The server is idle. Why are new connections timing out?Differentiate incomplete SYN requests from completed connections awaiting accept, measure burst capacity and repair slow accept loops.
- Part 4 of 6
Staff
All services are healthy. Who used the last outbound port?Calculate destination-specific NAT mapping occupancy and distinguish gateway port exhaustion from local ephemeral ports and bandwidth limits.
- Part 5 of 6
Staff
Tiny requests work. Why do large responses hang over the VPN?Diagnose size-dependent VPN loss through tunnel overhead, MTU and MSS arithmetic, ICMP feedback and black-hole-safe probing.
- Part 6 of 6
Principal
The load balancer says healthy. Why do 30 percent of requests fail?Detect partial backend failures invisible to liveness probes using client-outcome dimensions and safely bounded outlier ejection.