<< back to Guides
📊 System Design: Capacity Estimation (Deep Dive)
Capacity estimation helps you predict the resources your system will need to handle expected load, prevent downtime, and plan for scaling.
🎯 Why Capacity Estimation Matters
- Avoids underprovisioning (crashes) and overprovisioning (wasted cost)
- Helps choose storage engines, compute resources, and databases wisely
- Provides confidence in meeting SLAs and SLOs
- Sets a solid baseline for autoscaling or load balancing
🧠 Inputs to Estimate
-
Traffic Load
- Requests per second (RPS) or queries per second (QPS)
- Daily Active Users (DAU) / Monthly Active Users (MAU)
- Peak load vs average load
-
Data Size
- Average payload size per request/response
- Daily data growth rate (e.g., logs, user uploads, analytics)
-
Retention Requirements
- How long data needs to be stored (e.g., 30 days, 6 months, forever)
-
Processing Overhead
- CPU usage per request
- Memory footprint per user/session
-
Network Throughput
- Bandwidth required for serving requests
- Ingress/Egress load between services
🔍 Example Questions to Ask
- What is the expected RPS at launch?
- What’s the peak traffic multiplier? (e.g., 5x daily average)
- How big is each write or read operation in bytes?
- How long do we store user logs, uploads, or sessions?
- How many database read/write ops per second are expected?
📐 Estimation Formula Examples
📦 Storage Estimation
Daily Storage = Requests per Day × Payload Size
Total Storage (N days) = Daily Storage × Retention Period
Example:
1,000,000 requests/day × 2 KB = ~2 GB/day
Retention = 90 days → 180 GB total
🧮 Compute Estimation
Requests per second = Total Requests per Day / 86,400
If each request takes 20ms of CPU:
CPU Load = RPS × 20ms
= 1000 RPS × 20ms = 20 CPU cores
Add headroom (~30%) → 26 cores
🔄 Database IOPS Estimation
- Read Ops = RPS × Read ops/request
- Write Ops = RPS × Write ops/request
Add 50% overhead for replication/indexing
⚙️ Tools & Techniques
- Spreadsheets – Manual calculations for traffic, storage, CPU
- Load Testing Tools – Apache JMeter, k6, Locust
- Monitoring Baselines – Use real metrics if available (Grafana, Datadog)
- Capacity Planning Sheets – Structured templates to track estimates
🧱 Capacity Estimation Framework
1. Estimate baseline RPS / QPS
2. Factor in peak load (e.g., 2x or 10x burst traffic)
3. Estimate per-request data and processing size
4. Project data growth (daily, monthly, yearly)
5. Estimate backend capacity (DB IOPS, bandwidth)
6. Add buffer (25–50%) for safety and growth
💡 Pro Tips
- Always plan for peak, not average.
- Use real usage data from similar apps when possible.
- Leave room for future features and traffic spikes.
- Validate assumptions with engineers and business teams.
- Revisit estimations as product evolves or usage patterns shift.