Compute Troubleshooting and Optimization
Explore how to diagnose and optimize AWS EC2 performance bottlenecks by analyzing CPU, memory, disk, and network metrics. Understand a systematic troubleshooting approach that helps form evidence-based hypotheses, execute small reversible changes, and verify improvements to maintain stable, cost-effective application performance in production environments.
A user reports the service is slow, and the symptom looks identical across three incidents. In the first, CPUUtilization sits near the ceiling while request latency climbs. In the second, CPU looks fine, but free memory collapses and the OS spends time swapping, so latency spikes even though the app code did not change. In the third, CPU and memory look normal, yet requests stall because EBS throughput is saturated and disk queueing grows.
During an incident there is often only one safe change window, so guessing is expensive. The reliable pattern is a repeatable loop that turns a slow service report into one bottleneck hypothesis, one bounded adjustment, and a verification pass against the same baseline metric.
The troubleshooting loop
The core loop stays the same whether the goal is restoring performance or tuning cost. Because a single hypothesis is only as good as the evidence behind it, the loop starts with a baseline from the moment users feel pain, then pulls host metrics and volume or network metrics that can explain the slowdown, before forming a single hypothesis about the primary constraint.
Evidence is strongest when it comes from the resource that is ...