Network & Infrastructure / Telemetry, Capacity & Incident Analysis
Capacity percentiles and the failure-state demand hidden by averages
A circuit is sized from its normal 95th-percentile utilization, but overloads when a second circuit fails. Steady-state percentiles omit traffic redistribution and recovery demand.
Environment and inputs: Network and infrastructure observability; Prometheus examples describe metric semantics rather than a universal exporter schema. Diagnostic/design example; not tested on production equipment.
Collect the evidence
Measure demand by path and traffic class, then model each required failure scenario. Include backups, replication, growth and correlations between peaks rather than summing unrelated averages blindly.
Explain the result
A percentile summarizes an observed population under specific conditions. It does not guarantee capacity for a topology that has fewer surviving links or a different traffic distribution.
Plan the fix
Plan usable capacity against the agreed degraded-mode objective and reserve, documenting any intentional reduction in service during failure. Validate shaping and application behavior at the surviving bottleneck.
Check the result and rollback
Exercise a bounded path-loss scenario and compare queues, latency and useful throughput. Revert consolidation if the remaining capacity cannot satisfy the approved failure-state workload.