System Design Numbers to Know

Architecting scalable systems requires standard latency, bandwidth, and hardware numbers to perform back-of-the-envelope estimations.

1. Latency Numbers (Jeff Dean)

Understanding compute and network limits is essential for justifying caching and replication choices:

L1 Cache Reference: 0.5 ns
L2 Cache Reference: 7 ns
Main Memory Access: 100 ns
SSD Random Read (Seek): 16,000 ns (16 µs)
Read 1 MB sequentially from RAM: 250,000 ns (250 µs)
Read 1 MB sequentially from SSD: 1,000,000 ns (1 ms)
Round-trip within same Data Center: 500,000 ns (0.5 ms)
WAN Round-trip (US to Europe): 150 ms

2. Bandwidth & Storage Rules

• Network Bandwidth: Standard server links handle 1 Gbps to 10 Gbps.
• SQL Server: Handles ~1,000 to 5,000 write QPS depending on schema complexity.
• NoSQL Server (Cassandra/DynamoDB): Scales horizontally to 10,000+ write QPS per node.
• Redis Instance: Handles ~100,000 read queries/sec in memory.

3. References & Tech Blogs

• Interactive Latency Numbers Every Programmer Should Know
• Google Systems: Back-of-the-Envelope Estimation Techniques