We should get rid of average CPU utilization

# Summary Average CPU utilization is a misleading metric for performance monitoring because it masks critical latency issues, particularly in latency-sensitive applications. The author describes debugging a production timeout problem where CPU utilization appeared normal, but the container was being throttled by Linux's CFS (Completely Fair Scheduler) due to resource limits, a phenomenon invisible to standard monitoring tools. The article argues that average CPU utilization only answers a cost question and fails to reveal how workloads actually experience delays, where even small increases in utilization can cause exponentially longer wait times.