Latency Numbers Every Engineer Should Know
What you'll learn: The essential latency benchmarks that help you reason about system performance and make informed design decisions.
Why These Numbers Matter
When designing systems, you can't optimize what you can't measure. Certain latency values appear so frequently in system design that memorizing them becomes a superpower. They let you quickly estimate whether your design is feasible and where bottlenecks might appear.
The Core Benchmarks
Think of these as the "speed limits" of different operations in computing:
Memory Access (~100 nanoseconds)
Reading from RAM is incredibly fast—like grabbing a book from your desk.
Disk Seek (~10 milliseconds)
Finding data on a spinning hard drive takes 100,000x longer than memory access—like walking to a library across town instead of grabbing that book from your desk.
SSD Random Read (~150 microseconds)
Solid-state drives are faster than spinning disks but still ~1,000x slower than RAM—like driving to that library instead of walking.
Network Round-Trip Within Data Center (~0.5 milliseconds)
Sending data between servers in the same building is reasonably fast but still millions of times slower than memory access.
Network Round-Trip Cross-Continent (~150 milliseconds)
Sending data across the Atlantic and back takes 300x longer than within a data center—the speed of light becomes your enemy.
Practical Impact
These numbers immediately tell you that:
- Memory operations are essentially "free" compared to disk or network
- Network calls dominate your latency budget
- Reading from disk should be minimized or batched
- Geographic distance matters significantly for user experience
Key Takeaway: Memorizing these latency benchmarks lets you quickly identify performance bottlenecks and make order-of-magnitude estimates during system design discussions—memory is nanoseconds, disk is milliseconds, network within a datacenter is sub-millisecond, and cross-continent is hundreds of milliseconds.