Application Performance Tuning

Honest latency baselines, profilers you can read, budgets that fail the build, and the routines and targeted fixes that make slow pages, queries and services fast and keep them fast.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Then open the Marketplace in the app, search for Application Performance Tuning, and click Install. It goes into your own files, and works offline from then on.

What it installs.

1

Pillar

1

Area

50

Projects

201

Starter tasks

⚙️ Software Engineering

Application Performance Tuning

Profiling and speeding up slow pages, queries and services, setting performance budgets and keeping them from regressing over time.

  1. Inventory of slow pages, queries and services
  2. Latency baseline at p50, p95 and p99
  3. Choosing the speed metrics for each surface
  4. Profiler toolkit for every production runtime
  5. Slow query log switched on and first review
  6. Quiet benchmark environment that gives repeatable numbers
  7. Performance budgets for key pages and endpoints
  8. Real user page speed measurement in the field
  9. Team working agreement for performance changes
  10. Weekly slowest endpoints review
  11. Performance budget checks in the build pipeline
  12. Microbenchmark regression suite for hot code paths
  13. Monthly slow query review
  14. Monthly field page speed review
  15. Performance experiment log
  16. Quarterly performance report for the team
  17. Before and after comparison for every release
  18. Performance issue triage queue
  19. Reading flame graphs fluently
  20. Latency percentiles and tail latency maths
  21. Query execution plans with EXPLAIN
  22. Memory profiling and leak hunting
  23. Browser performance panel and the critical rendering path
  24. Benchmarking without fooling yourself
  25. Lock contention and concurrency diagnosis
  26. Garbage collector behaviour in your runtime
  27. N+1 query elimination across the ORM
  28. Response caching for one expensive read endpoint
  29. Image and media weight cut on key pages
  30. JavaScript bundle cut for the landing page
  31. Connection pool sizing for database and HTTP clients
  32. Moving slow work off the request path
  33. Startup and cold start time reduction
  34. Optimise or rewrite decision for one hot path
  35. Database index and query rewrite pass
  36. Pre-launch performance readiness review
  37. Performance comparison for a runtime or framework upgrade
  38. Pre-peak hot path tuning sprint
  39. Dedicated performance fix week
  40. Speed and business outcome case for stakeholders
  41. Solo engineer's minimum performance routine
  42. Rotating performance champion on the team
  43. Inheriting a slow codebase you did not write
  44. Tuning a small product on one modest server
  45. Speed on low-end phones and slow networks
  46. Performance skills map for the team
  47. Cache-friendly data layout in a hot loop
  48. Tail latency reduction for a fan-out service
  49. Database server configuration tuning
  50. Allocation-free hot path in a latency-critical service

Where it starts.

Each project arrives with its purpose, milestones and first tasks written out. Here is how the first three read.

Inventory of slow pages, queries and services

Performance work usually starts from one loud complaint and ends up fixing whatever the loudest person noticed. Collecting every known slow spot from support tickets, monitoring, the error tracker and the team's own grumbles gives you a ranked list where effort can go to the paths that cost users the most time.

First tasks

  • Search the last 90 days of support tickets for words like slow, timeout and spinning
  • Export the 20 slowest endpoints by p95 from the monitoring tool
  • Multiply each path's latency by its daily request count to estimate time lost
  • Publish the ranked list and name an owner for it

Latency baseline at p50, p95 and p99

Averages hide the users who wait longest, so a service with a 120 ms mean can still leave one request in a hundred waiting three seconds. Recording the median, the 95th and the 99th percentile for each key endpoint and page over a normal week gives every later change a number to beat and an honest picture of what users feel.

First tasks

  • Pick the ten to fifteen paths users and revenue depend on most
  • Confirm the monitoring tool stores histograms rather than pre-averaged values
  • Record p50, p95 and p99 for each path over a normal week
  • Note the traffic level and any incidents during the baseline week

Choosing the speed metrics for each surface

Web pages, JSON APIs, nightly batch jobs and mobile screens are slow in different ways, and one metric cannot describe them all. Deciding up front that pages are judged on Largest Contentful Paint and Interaction to Next Paint, APIs on percentile latency, and jobs on throughput and completion time stops arguments later about whether something actually got faster.

First tasks

  • List every kind of surface the product exposes to users or other systems
  • Read the current Core Web Vitals definitions and thresholds for the page metrics
  • Choose a primary and secondary metric for each surface type
  • Record the choices in the engineering guide and share them in the team channel

Make it yours.

Install Application Performance Tuning, then tell Aurum the date you're aiming for. It checks whether the plan fits the weeks you actually have.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Blueprints are general information to help you organise your own work. They are not medical, financial, legal or other professional advice. See the Terms.