Detailed Design of Bigtable: Part II

Many tables are kept in a Bigtable cluster. A table in Bigtable is made up of tablets, each of which stores all the data related to a specific range of rows. Each table starts out with only one tablet. As a table expands, it is automatically divided into many tablets, each of which has a standard size of between 100 and 200 MB. Let’s look at how we can locate and assign tablets and how read/write works in Bigtable.

How to locate the tablets

As tablets can migrate from server to server due to load balancing, tablet server failures, and so on, how do we figure out the right tablet server given a row? To find the answer to this question, we must locate the tablet whose row range includes the target row. To save tablet location data, Bigtable uses a three-level structure similar to that of a B+ treeBplustree.