💬
✉️
📸
▶️

Loading…

Capacity planning: turn users into requests, storage, and bandwidth

Capacity Planning

Capacity Planning

3 of 5

Storage Estimation

Tiny objects × billions of users = numbers that will surprise you.

A single tweet is tiny — 280 characters, about 350 bytes. But Twitter gets 500 million tweets *per day*. That tiny object, multiplied by that many users over time, is how you end up needing an exabyte-scale object storage system. The formula is always the same.

Formula

Daily storage = writes/day × avg object size × replication factor

writes/dayNew objects created every day (posts, messages, uploads)
avg object sizeText: ~300B · Image: ~200KB · Video: ~50MB
replication factor3× standard (1 primary + 2 replicas for durability)

Use decimal units here: 1 KB = 1,000 bytes, 1 GB = 10⁹ bytes, 1 TB = 10¹² bytes, 1 PB = 10¹⁵ bytes. Binary units (KiB, GiB, TiB) use powers of 1,024. Calculate raw first, then multiply by replication and retention days.

Worked exampleTwitter: 500M tweets/day with media
1
Text storage500M × 350 bytes
=175 GB/day

Tiny! Text is almost free.

2
Media50M photos (10%) × 200 KB
=10 TB/day

200,000 ÷ 350 ≈ 571: one image is about 571× a tweet in bytes

3
Total raw175GB + 10TB
=~10.2 TB/day

Media dominates. Text barely registers.

4
With replicas10.2 TB × 3
=30.6 TB/day

Durability isn't free

5
Yearly30.6 TB × 365
=~11 PB/year

At $0.023/GB on S3, that's ~$250,000/month

6
5 years11 PB × 5 (+ growth)
=~70+ PB

This is why Twitter built their own object storage

Takeaway: You can't store 70 PB affordably on a general cloud. The math forces you into CDN + tiered storage + compression + object lifecycle policies.