High-Scale Distributed Unique ID Generation: Snowflake, UUIDv7, and Ticket Servers

Across distributed databases featuring thousands of sharded tables, central auto-increment sequences create an insurmountable choke point. Conversely, adopting randomly generated UUIDv4 strings shatters B-Tree page locality in PostgreSQL and MySQL InnoDB, leading to massive write amplification and index fragmentation.

1. The Twitter Snowflake 64-Bit Architecture

The Snowflake algorithm produces k-sortable 64-bit integer identifiers with clear bit partitioning:

• 1 bit: Unused sign bit (fixed at 0).
• 41 bits: Epoch Millisecond Timestamp (providing ~69 years of lifespan).
• 10 bits: Node/Machine ID (supporting up to 1024 unique generator nodes).
• 12 bits: Sequence Number (enabling 4096 unique IDs per millisecond per node).

Throughput exceeds 4 million IDs per second per machine with zero network round trips or inter-node locking.