Back-of-the-Envelope Estimation: Bandwidth and Bottlenecks
So far, we’ve learned: how to estimate requests (QPS) how to estimate storage (GB/day) Now we answer a deeper question: How much data is moving through the system every second? This is where bandwidth estimation
So far, we’ve learned:
- how to estimate requests (QPS)
- how to estimate storage (GB/day)
Now we answer a deeper question:
How much data is moving through the system every second?
This is where bandwidth estimation comes in.
🚀 What is Bandwidth?
Bandwidth means:
how much data flows through your system per second
It’s usually measured in:
- KB/sec
- MB/sec
- GB/sec
🧠 The Core Formula
Bandwidth = QPS × Data per Request
This connects everything we’ve learned so far.
🔢 Basic Example
Let’s say:
- 100 requests per second
- each request = 2 MB
Then:
100 × 2 MB = 200 MB/sec
So your system needs to handle:
200 MB of data every second
⚡ Why Bandwidth Matters
Even if your system handles QPS well:
- large payloads can overload the network
- slow responses increase latency
- users experience delays
So bandwidth is often a hidden bottleneck
🧩 Real Intuition
Think of each request having “weight”:
- text request → light (KB)
- image upload → medium (MB)
- video streaming → heavy (GB)
So bandwidth is:
“how heavy your traffic is per second”
🚨 What is a Bottleneck?
A bottleneck is:
the first resource that fails under load
In most systems, you check:
- CPU (can it handle QPS?)
- Network (can it handle bandwidth?)
- Storage (can it handle data growth?)
🔍 Example: Finding the Bottleneck
Let’s say your system has:
- 1000 QPS
- 10 KB per request → ~10 MB/sec bandwidth
And your system capacity is:
- CPU: 2000 QPS
- Network: 5 MB/sec
Step 1: Check CPU
- Required: 1000 QPS
- Capacity: 2000 QPS → OK
Step 2: Check Network
- Required: 10 MB/sec
- Capacity: 5 MB/sec → NOT OK
👉 Conclusion
Network is the bottleneck
Even though CPU is fine.
🧠 Key Insight
A bottleneck is NOT the biggest number.
It is:
the first constraint that gets exceeded
🔗 Connecting Everything
Now you can analyze any system:
- QPS → request load
- Storage → data growth
- Bandwidth → data flow
- Bottleneck → system limit
⚠️ Common Mistakes
❌ Ignoring payload size
QPS alone is not enough.
❌ Assuming CPU is always the problem
Often, network or storage fails first.
❌ Not comparing with capacity
You must always compare:
demand vs system limits
🧠 Final Takeaway
Back-of-the-envelope estimation is not about numbers.
It’s about answering:
“What will break first in this system?”
👉 What’s next?
In the final part, we’ll apply everything to real systems:
- URL shortener
- messaging system
- feed system
If this helped, share your thoughts 👇
Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.