Latency Numbers Every SRE Should Know

nanosecond = 1/1,000,000,000 second microsecond = 1/1,000,000 second millisecond = 1/1000 second

Sub-Nanosecond Range

• Accessing CPU registers
• CPU Clock Cycle

1-10 Nanosecond Range

• L1/L2 cache
• Branch Misprediction in CPU pipelining

10-100 Nanosecond Range

• L3 cache
• Apple M1 referencing main memory (RAM)

100-1000 Nanosecond Range

• System call on Linux
• MD5 hash a 64-bit number

1-10 Microsecond Range

• Context switching between Linux threads

10-100 Microsecond Range

• Process a HTTP request
• Reading 1 megabyte of sequential data from RAM
• Read an 8k page from an ssd

100-1000 Microsecond Range

• SSD write Latency
• Intra-zone networking round trip in most cloud providers
• Memcache/Redis get operation

1-10 Millisecond Range

• Inter-zone networking Latency
• Seek time of a HDD

10-100 Millisecond Range

• Network round trip between US-west and US-east coast
• Read 1 megabyte sequentially from main memory

100-1000 Millisecond Range

• Some encryption/hashing algorithms
• TLS handshake
• Read 1 Gigabyte sequentially from an SSD

1 second+

• Transfer 1GB over a cloud network within the same region