Borg System Architecture

Borg System

• A Borglet is marked as down if it doesn't respond to several poll messages, and its tasks are rescheduled on other machines.
• When communication is restored, the Borgmaster instructs the Borglet to kill rescheduled tasks to prevent duplicates.

Borgmaster Architecture

• A Borgmaster consists of two processes: the main Borgmaster process and a separate scheduler.
• The main Borgmaster process handles client RPCs, manages state machines, communicates with Borglets, and provides a web UI.
• The scheduler is responsible for scheduling tasks and operates on a cached copy of the cell state.

Scalability

• A single Borgmaster can manage thousands of machines in a cell, with arrival rates above 10,000 tasks per minute.
• A busy Borgmaster uses 10-14 CPU cores and up to 50 GiB RAM.

Borg Architecture

• A Borg cell consists of a set of machines, a Borgmaster, and an agent process called the Borglet that runs on each machine.
• All components of Borg are written in C++.

Borglet Operation

• A Borglet continues normal operation even if it loses contact with the Borgmaster, ensuring currently running tasks and services remain up.

Fauxmaster

• Fauxmaster is a high-fidelity Borgmaster simulator that can read checkpoint files and contains a complete copy of the production Borgmaster code.
• It is useful for capacity planning, sanity checks, and debugging failures.