Milvus — Architecture Introduction

The architecture of Milvus, a robust vector database for machine learning, is designed to handle heavy data loads efficiently. The following diagram is from office Milvus documentation:

Pic from milvus

If you feel the above diagram is a bit complicated, here is a simplified version:

Decoupled Storage and Compute

A key feature of Milvus is its decoupled storage and compute architecture. This design separates the storage and processing of data, enabling each to scale independently based on workload demands.

In traditional monolithic systems, the computational resources (processors, RAM) and storage resources (hard drives, SSDs) are tightly coupled. They exist on the same machine and scale together. If you need more computing power, you typically need to upgrade your storage capacity as well, even if you don’t require the extra storage.