Large-scale performance
On this page 9 topics
Overview
Topics
- Capacity forecasting
- workload modeling
- cost-performance optimization
Learn first
3.5 h of practice (estimate)
Learning outcomes
- Forecast capacity across many services from business drivers.
- Build workload models that represent real traffic mixes, including new products (marketplace sellers).
- Optimize cost and performance together (cost per transaction at target latency).
- Set performance budgets for teams and track them.
Lessons
Lab and exercises
Lab
SRA-007-LAB1 · Three-year forecast
Outcomes: 1–4 · Time: ~2 h
- Collect business drivers for Commerce 2030: orders, sellers, listings, countries, campaigns.
- Measure or assume load ratios per driver and build a workload model including marketplace shapes.
- Produce a three-year capacity forecast with low, base, and high cases.
- Compute cost per order at target latency for each year and propose optimizations.
- Set performance budgets for three teams on the checkout path.
Acceptance: assumptions explicit; sensitivity to growth rate shown.
Exercises (case-based; six exercises)
| ID | Outcome | Task |
|---|---|---|
| SRA-007-EX01 | 1 | Orders double yearly. Given requests per order for six services, forecast peak requests per second for two years and identify the first bottleneck. |
| SRA-007-EX02 | 1 | Add sensitivity: show the effect of 1.5x and 3x annual growth on the forecast and on cost. |
| SRA-007-EX03 | 2 | Build a workload model for seller bulk uploads and inventory sync and state how you would validate it. |
| SRA-007-EX04 | 2 | Explain why a load test with only buyer traffic would miss risks at marketplace launch. |
| SRA-007-EX05 | 3 | Choose among three optimizations (cache, batching, right-sizing) by cost per order at target latency; justify the pick. |
| SRA-007-EX06 | 4 | Split an 800 ms checkout budget among five teams, define tracking, and describe what happens when a team exceeds its share. |
Show solutions
- EX01 Multiply orders by ratio per service, apply a peak-to-average factor and headroom; the service with the highest load-to-capacity ratio goes first, often the one that scales worst (state your numbers and assumptions).
- EX02 Run the same model with each rate; show how the bottleneck date and capacity spend move; highlight the decision that changes first.
- EX03 Rates by seller count and listing size, burst patterns (batch windows), payload sizes; validate with synthetic load tests, analog data from similar platforms, and early production seller data.
- EX04 Buyer traffic is read-heavy and spread; sellers add write bursts and indexing load, shifting bottlenecks to ingestion, search indexing, and storage.
- EX05 Compute savings and latency effect per option; pick the one with lowest cost per order meeting the target and with acceptable complexity; verify by measurement.
- EX06 Example: gateway 50 ms, pricing 100 ms, inventory 150 ms, payment 350 ms, remaining 150 ms; dashboards per team; exceeding triggers a plan or a budget renegotiation, not blame.
Outcome 4 is also covered by the lab budgets and the quiz.
Workplace evidence: a capacity forecast used in budgeting.
Quiz
Outcomes: O1 capacity forecasting · O2 workload modeling · O3 cost-performance optimization · O4 performance budgets.
A self-check for you: nothing is graded, saved or sent anywhere. Choose an answer, then check it. You can try again as often as you like.
Practical part
Six case exercises SRA-007-EX01 to EX06 plus the lab, reviewed as a capacity forecast. Pass: at least five of six exercises correct, quiz at least 8/10, and a forecast with explicit assumptions, growth sensitivity, and cost per order at target latency per year.
Common mistakes
Linear extrapolation; ignoring new traffic shapes from new products.
Assessment
Forecast review.
Workplace evidence
A capacity forecast used in budgeting.
Career relevance
Cost architecture (ARC-021).