Visualizing Back-of-the-Envelope Numbers

December 1, 2024

System Design

It can be difficult developing a intuition for the numbers thrown around in system design interviews. For my own sake, I thought I'd try to visualize them.

Data Type Sizes

First, a table of data type sizes. These are Postgres data types, but they should be similar for other systems. The squares are scaled in size logarithmically, which means that the differences are really much bigger than they seem here.

TypePostgres Data TypesSizeNotes
Integer
  • SMALLINT: 2 bytes
  • INTEGER: 4 bytes
  • BIGINT: 8 bytes
2 bytes
4 bytes
8 bytes
Used for numeric fields, IDs, and counters. Larger sizes support larger ranges.
Floating Point
  • REAL: 4 bytes
  • DOUBLE PRECISION: 8 bytes
4 bytes
8 bytes
Used for scientific calculations or approximate values.
Timestamp
  • TIMESTAMP: Without timezone
  • TIMESTAMPTZ: With timezone
8 bytes
Used for date and time fields, with or without timezone information.
UUID
  • UUID: 16 bytes
16 bytes
Universally unique identifiers for distributed systems.
Enum
  • ENUM: 1-4 bytes
1 byte
4 bytes
  • Represents a predefined set of values.
  • Examples: User roles, statuses, or category lists.
String
  • VARCHAR(n): Up to n characters
  • TEXT: Unlimited length
~100 bytes
~1 KB
~10 KB
  • Short: Usernames, titles, or short sentences.
  • Medium: Descriptions or paragraphs of text.
  • Long: Articles, blog posts, or large text fields.
JSONB
  • JSONB: Variable length
~100 bytes
~1 KB
~10 KB
  • Small: Simple JSON objects.
  • Medium: Nested JSON or API responses.
  • Large: Complex JSON with many fields.
BYTEA
  • BYTEA: Variable length
~1 KB
~100 KB
~1 MB
  • Small: Icons or thumbnails.
  • Medium: Profile pictures or compressed files.
  • Large: High-resolution images or large documents.
Legend:
1 byte
100 bytes
1 KB (1,024 bytes)
10 KB
100 KB
1 MB (1,048,576 bytes)

Time Scales

Second, this is a table of time scales. Some quick comparisons: it's 160x faster to read from RAM than do a random read from SSD. It's 4x faster to read 1MB sequentially from RAM than to do the same from SSD. Squares are scaled in size logarithmically.

OperationTimeVisualization
L1 Cache Reference
Fetching a hot variable in a running JavaScript function.		 0.5 ns
L2 Cache Reference
Accessing slightly less frequent in-memory data.		 7 ns
L3 Cache Reference
Resolving reused middleware configurations in Node.js.		 20-30 ns
Main Memory Reference
Fetching data from a Redis cache stored in RAM.		 100 ns
Compress 1KB of Data in Memory
Gzip compressing an HTTP response in Express.js.		 3,000 ns
Send 1KB over 1 Gbps Network
Sending a small JSON response over the network.		 10,000 ns
SSD Random Read
Querying a session ID from persistent Redis storage.		 16,000 ns
Read 1MB Sequentially from Memory
Iterating through a large in-memory buffer of user data.		 250,000 ns
Read 1MB Sequentially from SSD
Retrieving a 1MB file stored on disk for a static site.		 1,000,000 ns
Disk Seek
Accessing a log file stored on a spinning disk.		 10,000,000 ns
Send Packet US to Europe and Back
Calling an API hosted in a European data center.		 150,000,000 ns
Legend:
0.5 ns
1 ns
100 ns
1,000 ns
10,000 ns
1,000,000 ns
10,000,000 ns
150,000,000 ns