What are Performance Metrics?#
Before you design anything — how do you know if your system is working well?
You measure it. Performance Metrics are the measurements that answer that question.
The Doctor Analogy#
When a doctor checks your health, they don't just say "you look fine". They measure: - Blood pressure - Heart rate - Temperature
Each number tells them something specific. A high temperature means infection. Low blood pressure means something else entirely. They can't treat you without knowing which number is off.
Your system works the same way. When something is wrong, you need to know which metric is failing before you can fix it.
How the metrics connect#
- Latency tells you how fast one request moves
- Throughput tells you how many requests move simultaneously — directly shaped by latency and threads
- Bandwidth tells you how much data those requests carry — a different bottleneck entirely
- Percentiles make all three measurable and honest — without them, every claim about latency or throughput is just a vibe
The mindset shift
Every architecture decision you make — adding a cache, sharding a database, using a queue, deploying a CDN — is a direct response to one of these metrics being bad.
Before you propose any solution in an interview, ask yourself: which metric am I fixing right now?
Why this matters in interviews#
Interviewers will ask you to define requirements. Vague answers fail. Specific metric-based answers pass.
❌ "The system should be fast" ✅ "P99 latency should be under 100ms"
❌ "It should handle a lot of traffic" ✅ "We need to support 50,000 requests per second at peak"