Common Pitfalls in Estimation
What You'll Learn
You'll discover the most frequent mistakes engineers make when performing back-of-the-envelope calculations—and how to avoid them.
Why Estimation Goes Wrong
Even when you understand QPS, storage, and bandwidth calculations, it's easy to produce wildly inaccurate estimates. Three pitfalls trip up beginners (and even experienced engineers):
Ignoring Replication
The mistake: You calculate storage as 1 million users × 1 KB per profile = 1 GB, then stop.
The reality: Production systems replicate data for availability and reliability. With 3-way replication (common for databases), your actual storage need is 3 GB, not 1 GB. Forgetting replication underestimates storage and costs by 3x or more.
Forgetting Overhead
The mistake: Treating theoretical capacity as usable capacity.
The reality: A server handling 10,000 QPS in theory won't sustain that in practice. You need headroom for peak traffic multipliers (remember those?), garbage collection, network retries, and system overhead. Rule of thumb: plan for 50-70% utilization at normal load. If you calculate needing 10 servers at 100% capacity, you actually need 15-20 servers.
Unrealistic Conversion Factors
The mistake: Using 1 KB = 1000 bytes when systems use 1 KB = 1024 bytes, or confusing bits with bytes in bandwidth calculations.
The reality: While these seem minor, they compound. Calculating CDN bandwidth as 100 Mbps when you meant 100 MB/s creates an 8x error—potentially costing hundreds of thousands in overprovisioning or causing catastrophic underprovisioning.
Real-World Impact
These aren't academic errors. Forgetting replication might mean running out of disk space at 3 AM. Ignoring overhead causes cascading failures under load. Wrong units bankrupt startups or crash launches.
Key Takeaway: Always account for replication multipliers, leave 30-50% overhead for real-world conditions, and double-check your units—bytes vs bits, 1000 vs 1024—before finalizing any estimate.