percentile
The value below which a given percentage of measurements fall; p99 latency is the time 99 in 100 requests beat.
In more detail
The p50, or median, shows a typical request and p99 exposes the slow tail. Averages hide the tail, which often reveals queueing, lock contention or a stalling dependency.
Examples
In AWS: The value below which a given percentage of measurements fall; the 99th percentile latency is slower than 99 per cent of requests.afternoon 50.5 48 48 48 98 # same mean, slower p99In the browser: Percentiles describe typical and bad experiences better than averages, which hide slow outliers.From 100 requests: mean 96 ms, median 40 ms, p99 1,500 ms. The p99 is not the worst case; the maximum was 2,400 ms.Percentiles, also called quantiles, describe the spread of a set of numbers better than an average does.In the course sample: mean 72 ms, median 40 ms, p95 59 ms, p99 nearly 1,400 ms.In this course: Percentiles describe what users actually experience better than a mean. Tools interpolate slightly differently, so name the tool and method in reports.Related terms
- tail latency — The latency of the slowest requests, measured by high percentiles such as p99.
- latency — The time between asking for something and getting the answer back.
- histogram — A metric that counts measurements, such as request durations, into ranges of values so their distribution and percentiles can be calculated.
- metric — A number measured over time, such as requests per second or error count, that shows how a system behaves in aggregate.
- alarm — A rule that watches a metric and changes state, and notifies someone, when the metric crosses a threshold for long enough.
- median — The middle value when all observations are sorted: half are faster and half are slower.
- mean — The sum of all values divided by how many there are; what people usually call the average.
- Core Web Vitals — Google's three key user-experience metrics for web pages: Largest Contentful Paint for loading, Interaction to Next Paint for responsiveness and Cumulative Layout Shift for visual stability.
- histogram_quantile() — The PromQL function that estimates a percentile from a histogram's cumulative buckets.
- outlier — A value far from the rest of a set of measurements, such as one incident that took days to contain when most took hours.
- service level objective — A target for a service level indicator over a time window, such as 99.9 per cent of checkouts succeeding over 30 days.
Taught in
- Choosing A Model — AI Engineering: LLM Apps, RAG & Agents
- Tracing And Monitoring — AI Engineering: LLM Apps, RAG & Agents
- Capstone: The Tasks Assistant — AI Engineering: LLM Apps, RAG & Agents
- Load Testing — Modern API Architecture: REST, GraphQL & gRPC
- Asking Telemetry Why, With KQL — Microsoft Azure DevOps Engineer Expert (AZ-400) Exam Prep
- Observability with CloudWatch — Cloud Infrastructure & Hyperscaler Architecture
- Metrics, Logs, and Alarms — Cloud Infrastructure & Hyperscaler Architecture
- Six Big Servers Doing Very Little: Rightsizing — CompTIA Cloud+ (CV0-004) Exam Prep
- Slow At Lunchtime: Logs, Traces And Alerts — CompTIA Cloud+ (CV0-004) Exam Prep
- Percentiles Rather Than Averages — DevOps & Site Reliability Engineering
- Service Level Objectives — DevOps & Site Reliability Engineering
- Core Web Vitals: Measuring And Optimizing LCP, INP, And CLS — Frontend Architecture & Web Platform
- Continuous Delivery: Blue-Green & Canary Pipelines — Production Full-Stack Engineering
- Slow, Or Broken? Metrics, Logs And Traces — Kubernetes and Cloud Native Associate (KCNA) Exam Prep
- The Festival, End To End: Final Review — Kubernetes and Cloud Native Associate (KCNA) Exam Prep
- Observability Versus Monitoring — Full-Stack Observability: Metrics, Tracing & Logging
- Metric Types: Summaries And Histograms — Full-Stack Observability: Metrics, Tracing & Logging
- Why A Histogram Beats An Average — Full-Stack Observability: Metrics, Tracing & Logging
- Calculating Percentiles With histogram_quantile — Full-Stack Observability: Metrics, Tracing & Logging
- Dashboards That Answer Questions — Full-Stack Observability: Metrics, Tracing & Logging
- Six Minutes Nobody Saw: Prometheus In One Outage — Prometheus Certified Associate (PCA) Exam Prep
- Ninety Minutes, No Question Count: Exam Strategy — Prometheus Certified Associate (PCA) Exam Prep
- Cart Versus cart: The Data Model And What A Scrape Returns — Prometheus Certified Associate (PCA) Exam Prep
- Fifteen Requests Nobody Saw: Aggregating Over Time — Prometheus Certified Associate (PCA) Exam Prep
- Fifteen Lines And No Answer: Aggregating Over Labels — Prometheus Certified Associate (PCA) Exam Prep
- Under A Second, On Average: Histograms — Prometheus Certified Associate (PCA) Exam Prep
- PromQL Sprint: Ten Queries At Minute Twenty Two — Prometheus Certified Associate (PCA) Exam Prep
- Thirty Second Checkouts Nobody Saw: Client Libraries — Prometheus Certified Associate (PCA) Exam Prep
- Instrumentation Sprint: Sixteen Calls On Code And Exporters — Prometheus Certified Associate (PCA) Exam Prep
- Twelve Green Panels: Dashboarding Basics — Prometheus Certified Associate (PCA) Exam Prep
- Seventy Seven Percent, Book Now? The Full Mock — Prometheus Certified Associate (PCA) Exam Prep
- One Week, Three Objectives: Final Review — Prometheus Certified Associate (PCA) Exam Prep
- SOC Operating Models, Tiers & Key Performance Metrics — Security Operations & Threat Hunting (SOC)
- Requirements: The Constraints That Shape The System — Scalable System Design & Distributed Architecture
- Latency Numbers Every Engineer Should Know — Scalable System Design & Distributed Architecture
- Observability As A Design Input — Scalable System Design & Distributed Architecture
- Capacity Planning As A Design Input — Scalable System Design & Distributed Architecture
- Load Testing: Percentiles Over Means — Automated Testing, TDD & Quality Engineering
References
- Google SRE book: Service level objectives
- Prometheus: Histograms and summaries
- CloudWatch statistics definitions: percentiles
- NIST/SEMATECH e-Handbook: percentiles
- Python statistics module: quantiles
Certification courses
- AWS AIP-C01 course and practice exam: AI engineering with LLM apps, RAG and agents
- AZ-400 course and practice exam: Microsoft Azure DevOps Engineer Expert
- CompTIA Cloud+ course and practice exam: CV0-004
- KCNA course and practice exam: Kubernetes and Cloud Native Associate
- PCA course and practice exam: Prometheus Certified Associate