Tablet Architecture and Distribution
What you'll learn: How BigTable splits table data into tablets (row ranges) and distributes them across tablet servers for horizontal scalability.
What Are Tablets?
In BigTable, tables don't live as one giant block of data. Instead, BigTable automatically partitions each table into tablets — contiguous ranges of rows sorted by row key. Think of a phonebook split into volumes: A-F, G-M, N-Z. Each volume is a tablet.
A tablet contains all columns for rows within its range. For example, one tablet might hold rows from user:0000 to user:1999, while another holds user:2000 to user:3999.
How Tablets Are Distributed
Tablet servers are the workers that manage tablets. Each tablet server is responsible for typically 10-1000 tablets. When you read or write data, your request routes to the tablet server currently serving the tablet containing that row key.
Tablets are not permanently tied to specific servers. BigTable's master server assigns tablets dynamically. If a tablet server becomes overloaded or fails, tablets can be reassigned to other servers. This provides both load balancing and fault tolerance.
The size of tablets adapts automatically: when a tablet grows too large (typically 100-200 MB), BigTable splits it into two smaller tablets. These can then be distributed independently.
Why This Matters
This architecture enables BigTable to scale horizontally. Need more capacity? Add more tablet servers, and tablets redistribute automatically. Hot row ranges? BigTable can split busy tablets and spread the load. You get elastic scalability without manual repartitioning.
Key Takeaway: BigTable partitions tables into tablets (sorted row ranges), distributes them across tablet servers, and dynamically rebalances them for scalability and fault tolerance.