Network Performance Diagnostic
Contention and capacity produce identical symptoms — slow applications, latency spikes, degraded performance at peak. The difference is what fixes them. Adding bandwidth to a capacity problem works. Adding bandwidth to a contention problem raises the threshold temporarily and the cycle repeats. This diagnostic helps you determine which one you’re dealing with.
Check every sign that applies to your organization. If you check three or more, you have a contention problem — and the result will tell you the four questions worth taking back to your team.
Each bandwidth addition raises the contention threshold — it doesn't change which workloads share which paths. The problem recurs at a higher load because the architecture hasn't changed. Only the ceiling has.
A network at 40% average utilization can be fully saturated during 15–20 minute peak windows when workload spikes overlap. Average utilization data completely flattens out these events, masking your infrastructure's true burst tolerances. This is why your dashboards look fine while users are experiencing jitter.
Workaround culture masks the problem from the people who could authorize the fix. The cost shows up as engineering overhead, not as downtime. Organizations with frequent degradation report 16x the total incident cost of organizations with rare outages — because degradation is harder to detect, attribute, and resolve than a clean outage. (Catchpoint, 2025)
Misdiagnosis is expensive before a single remediation step begins. Engineering hours ruling out application bugs and server load account for 20–40% of total incident cost before any fix starts. (Ponemon Institute, 2024)
The gap isn't rigor — it's scope. Per-workload review is the right process for evaluating individual workloads. It wasn't designed to model correlated demand. Without a system-level concurrency model, every new workload approved onto shared infrastructure carries unmodeled risk.