Back-of-the-Envelope: The Art of Rapid System Estimation

Step 2: Calculate Queries Per Second (QPS)

• Total Daily Requests = DAU × average actions per user.
• Average QPS = Total Daily Requests / 86,400 (seconds in a day, often rounded to 100,000 for ease of calculation).
• Peak QPS = Average QPS × 2 (or up to × 5, depending on whether the product experiences "golden hour" traffic spikes).

Step 3: Estimate Data Size Memorize standard data footprints:

• 1 Character = 1 byte (ASCII) or 2 bytes (Unicode).
• 1 User ID / Integer = 4 or 8 bytes.
• 1 Short text string / Metadata = 100 - 500 bytes.
• 1 Standard image = 100 KB - 2 MB.
• 1 Short video = 10 MB - 50 MB.

Step 4: Estimate Storage & Bandwidth

• Storage = (New data entities generated / day) × Average data size. Extrapolate this to 1-year and 5-year trajectories.
• Bandwidth = (Requests / second) × Average payload size.

4. Real-world Case Study

Context: Shopee wants to launch "Shopee Video Reviews," allowing buyers to upload product review videos (maximum 1 minute long). You need to calculate the storage and bandwidth requirements.

• App DAU: 50 million.
• Daily conversion rate: 5% of DAU make a purchase = 2.5 million orders.
• Video review rate: 10% of orders result in a video upload = 250,000 videos uploaded per day.
• Average video size: 20 MB (compressed).
• Read/Write ratio: 20:1 (Each video is viewed an average of 20 times per day).

• Daily new storage: 250,000 videos × 20 MB = 5,000,000 MB = 5 TB / day.
• 1-Year storage: 5 TB × 365 = ~ 1.8 PB / year. (Insight: This is a massive volume. The system will necessitate a Cold Storage architectural pattern for older, rarely viewed videos to optimize infrastructure costs).

• Ingress (Upload bandwidth): 5 TB / day divided by ~100,000 seconds = 50 MB/second.
• Egress (Download/Read bandwidth): With a 20:1 Read/Write ratio, download bandwidth is 20x higher = 50 MB × 20 = 1,000 MB/second (or 1 GB/second). (Insight: Sustaining 1 GB/second of continuous outbound traffic requires integrating a robust Content Delivery Network (CDN) to offload pressure from the origin servers).

5. Anti-patterns & Pitfalls (Trade-offs)

While BOTE is exceptionally powerful, professionals often fall into systemic traps:

• False Precision: Attempting to calculate highly specific results like "3,456.78 GB." BOTE is strictly about Orders of Magnitude. Round up or down aggressively to keep the math head-friendly.
• Ignoring Peak Load: Designing a system solely for Average QPS is a fatal error. A midnight Flash Sale can spike traffic 10x to 50x above the baseline. If your BOTE omits a Peak Load multiplier, the system will crash on launch day.
• Losing Sight of Business Logic: Blindly applying formulas without considering product behavior. For instance, estimating RAM to cache 100% of the video database is a massive waste of capital. Applying the Pareto principle (80/20 rule) dictates that 20% of trending videos will drive 80% of the traffic; therefore, you only need to calculate cache memory for that 20%.
• Silent Assumptions: Stating a final conclusion like "We need 10 servers" in a meeting without walking through your DAU and conversion assumptions will instantly destroy your credibility. Always narrate your variables.

Ready to master system estimation?

Put this framework into practice today. Select a feature currently sitting in your backlog, write down your DAU assumptions, quantify the payload size, and attempt to mentally calculate the baseline QPS in under 5 minutes. True system thinking begins the moment you understand your product's physical limits.