Git at Every Scale

Git is a distributed version control system designed by Linus Torvalds for Linux kernel development. Although it has become the industry standard, most enterprises rely on centralized hosting, making large-scale repository management still a challenging task.

The core difficulty in Git hosting lies in its packfile-based design. This binary file, which compresses code and metadata, is efficient locally but significantly limits server-side scalability and availability. Repositories need to be distributed across multiple disks and machines to enable parallel processing and fault tolerance, but this is difficult to implement on a single file system.

Three approaches exist as solutions: file system distribution, packfile distribution, and Git itself distribution. Since Git is a content-addressed data store, it might seem to map well to a distributed key-value store, but the actual repository structure is a DAG (Directed Acyclic Graph). Retrieving objects requires traversing the DAG step by step, so a simple key-value approach alone does not work.