Performance Metrics Vocabulary Exercises

1 / 5

A monitoring dashboard shows: "The p99 latency for the checkout API is 2.3 seconds."

What does "p99 latency" mean?

Percentile latency is more meaningful than average because it captures the tail of the distribution — the slowest requests that real users actually experience. A fast average can hide a badly slow 1%.

Percentile	Meaning
p50	Median — 50% of requests are faster than this
p95	95% of requests complete faster than this — only 5% are slower
p99	99% of requests complete faster — captures "tail latency" affecting 1% of requests
p999	99.9th percentile — used for high-traffic services where 0.1% still = many users

Key vocabulary: tail latency, percentile distribution, mean vs. percentile (mean hides outliers), latency SLO defined at p99.

2 / 5

A performance engineer reports: "Throughput dropped from 8,500 RPS to 3,200 RPS after the deployment."

What does "throughput" measure in a system performance context?

Throughput measures volume capacity — how many requests the service handles per second. It is distinct from latency (time for a single request). A service can have low latency but low throughput (can't handle many concurrent requests) — a key performance trade-off.

Metric	Unit	Measures
Throughput	RPS, TPS (transactions per second)	Volume — how many requests handled per second
Latency	ms, seconds	Speed — how long a single request takes
Saturation point	RPS at which latency begins to rise sharply	The throughput ceiling of the system

3 / 5

A performance engineer says: "We established a performance baseline before starting the optimisation work."

What is a performance baseline?

A performance baseline is a pre-change measurement snapshot. Without it, you cannot quantify improvement ("we reduced p99 by 40%") or detect regression ("p99 has worsened since last deploy"). Baselines enable data-driven performance conversations.

Term	Meaning
performance baseline	Measured metrics snapshot before a change — the reference for before/after comparison
regression testing	Comparing post-change metrics against the baseline to detect performance regressions
baseline deviation alert	An alert triggered when current metrics diverge significantly from the established baseline
performance trending	Tracking metric movement over time to detect gradual degradation

4 / 5

An SRE writes in a post-mortem: "The service experienced performance degradation under load during peak hours."

Which explanation most precisely describes "performance degradation"?

Performance degradation means metrics worsen under load — this often follows a non-linear curve: latency increases slowly at low load, then steeply near the saturation point (the "knee of the curve"). The service remains functional but slower.

Term	Meaning
degradation curve	The shape of latency increase as load grows — initially linear, then steep
capacity knee	The inflection point where degradation becomes non-linear
saturation point	The load level at which the system is fully utilised and can no longer keep up
graceful degradation	The system continues operating at reduced quality rather than failing completely

5 / 5

A monitoring dashboard shows memory utilisation growing linearly from 2 GB to 8 GB over 6 hours without restart.

What does this pattern suggest?

Linear memory growth is the classic memory leak signature: allocations accumulate without being released. This eventually causes OOM (Out of Memory) errors and process crashes, or forces frequent restarts. Compare against healthy patterns:

Memory pattern	Interpretation
Flat line	Healthy — stable memory usage
Sawtooth pattern	GC cycles working correctly — allocate, collect, allocate, collect
Linear growth	Memory leak — objects retained that should be released
Sudden spike	Possible large object allocation or traffic burst — investigate

Key vocabulary: heap growth, GC pressure, heap dump analysis, retain vs. release.