DEV Community

Saras Growth Space
Saras Growth Space

Posted on

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 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
Enter fullscreen mode Exit fullscreen mode

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
Enter fullscreen mode Exit fullscreen mode

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 👇

Top comments (0)