Essential System Performance Metrics Every Ops Engineer Should Track

Introduction

In today's digital era, software systems must remain stable under high load and concurrency to provide a good user experience. Monitoring performance metrics is essential for understanding a system's health.

Metric Classification

Metrics can be grouped into four dimensions:

Infrastructure metrics : health of servers, network, storage, etc.

Application metrics : response time, throughput, error rate, concurrency.

User experience metrics : page load time, render time, interaction latency.

Business metrics : revenue, conversion rate, user growth, and other product‑related figures.

Detailed Metrics

Infrastructure Metrics

CPU utilization – percentage of CPU usage; high values may cause slow response or crashes.

Memory usage – percentage of memory used; excessive usage can lead to slowdown or crashes.

Disk space usage – percentage of disk capacity used; high usage may cause failures.

Disk read/write speed – measured in MB/s; low speed can degrade performance.

Network latency and bandwidth – speed and delay of data transmission; higher bandwidth and lower latency improve responsiveness.

Process count – number of running processes; too many can exhaust resources.

System load – average load over 1, 5, 15 minutes; high load indicates heavy task processing.

Database execution time – time spent executing SQL statements; helps identify slow queries.

Throughput – QPS (queries per second) and TPS (transactions per second).

Cache hit rate – monitors cache efficiency, aiding SQL performance.

Application Metrics

Request response time – time from request to response, measured in ms; shorter is better.

Throughput – number of requests processed per time unit (QPS or RPM); higher indicates better capacity.

Error rate – ratio of failed requests to total requests, expressed as a percentage; lower is more stable.

Concurrency – number of simultaneous requests; insufficient handling leads to latency.

User Experience Metrics

Page load time – total time for a page to load, including network, DNS, server response, and asset download.

Page render time – time from start of loading to visual presentation; affected by browser performance, JS complexity, CSS size, images.

Interaction response time – time between user action and UI response; critical for perceived performance.

Business Metrics

User metrics – daily new active users, active users, retained users; reflect growth and engagement.

Behavior metrics – PV (page views), UV (unique visitors), conversion rate; indicate traffic and marketing effectiveness.

Product metrics – revenue, profit, average revenue per user, product ranking; measure commercial value.

Other Metrics

Middleware metrics – e.g., MQ, Nacos, JVM.

Stability metrics – e.g., “four nines”, “five nines” availability.

Reliability metrics – backup/recovery, cluster reliability.

Overall, the article emphasizes that metric selection should be driven by leadership and data, as highlighted by the code snippet

领导驱动、数据驱动

.