# System Design Back-of-the-Envelope Calculations: QPS, Storage, Bandwidth & Latency Math

> How to run clean 3-minute capacity estimates in L5 Senior and L6 Staff system design rounds—and use those numbers to justify sharding, caching, and queues.

- **Author:** Former Amazon Software Engineer (Crush the FAANG Interview)
- **Category:** System Design Fundamentals
- **Updated:** 2026-11-20
- **Canonical URL:** https://crushthefaanginterview.com/blog/system-design-back-of-the-envelope-estimation-guide

## Key Takeaways
- 1 day = 86,400 seconds (~10^5 seconds). 1 million requests/day is ~12 RPS average; 100 million requests/day is ~1,150 RPS average.
- Never stop at arithmetic: every estimate must drive an architectural decision (e.g., 45,000 write QPS exceeds a single primary Postgres node -> partition by shard key).
- Size peak traffic at 3x–5x average QPS, and size Redis hot-set memory using the 80/20 Pareto rule over active daily records.

## 1. Fast Mental Math Shortcuts for Interviews

Bar Raisers do not care whether 500M / 86,400 equals 5,787 or 5,800. They care about order-of-magnitude accuracy using powers of 10 and powers of 2, completed in under three minutes.

| Conversion Rule | Exact Value | Interview Shortcut |
| --- | --- | --- |
| Seconds per day | 86,400 s | 10^5 s (100,000 s) |
| Seconds per month | 2.592 x 10^6 s | 2.5 x 10^6 s |
| Seconds per year | 3.1536 x 10^7 s | 3 x 10^7 s |
| 1 Million / day | 11.57 RPS | ~12 RPS avg (~35–50 RPS peak) |
| 100 Million / day | 1,157 RPS | ~1.2K RPS avg (~4K–5K RPS peak) |
| 1 Billion / day | 11,574 RPS | ~12K RPS avg (~40K–50K RPS peak) |

## 2. Hardware & Network Baselines That Anchor Your Design

Once you compute read QPS, write QPS, and 5-year storage, compare them against realistic production node envelopes to prove why a single box works—or why horizontal partitioning is mandatory:

| Component | Practical Single-Node Envelope | When to Scale Out |
| --- | --- | --- |
| Primary Relational DB (Postgres/MySQL) | 5K–15K simple ACID writes/s; 2–4 TB comfortable working set | Add read replicas for high read:write ratios; shard when sustained writes > 10K/s or DB > 4 TB. |
| Redis In-Memory Cache Shard | 80K–100K ops/s single-threaded; 32–64 GB RAM per shard | Cluster across hash slots when hot working set > 50 GB or QPS > 80K/s; guard against hot keys. |
| Kafka Broker Partition | 10–50 MB/s per partition; millions of msgs/s across cluster | Partition by entity_id to preserve per-entity ordering while scaling consumer groups. |
| Object Storage + CDN | Virtually unbounded capacity; 99.999999999% durability | Offload all media blobs (> 100 KB) out of databases via presigned upload URLs. |

## Frequently Asked Questions

### Should I always do back-of-the-envelope math at the start of a system design interview?

Spend 2–3 minutes computing QPS, storage growth, and hot-cache RAM right after clarifying functional and non-functional requirements, or calculate numbers inline right before choosing database sharding and cache tiers.

## Related Video & Interactive Courses
- [Back-of-the-Envelope Calculations](https://crushthefaanginterview.com/course/sd-05-back-of-the-envelope-calculations)
- [Databases & Sharding](https://crushthefaanginterview.com/course/sd-09-databases)
- [Distributed Cache](https://crushthefaanginterview.com/course/sd-16-distributed-cache)
- [Distributed Rate Limiter](https://crushthefaanginterview.com/course/sd-19-rate-limiter)
