JFrog ML Architecture

The engineering behind JFrog ML machine learning platform

Overview

With JFrog ML, you can deploy and iterate on your machine learning models faster, gain control and visibility of your production machine learning environment, and ensure the privacy and security of your data and models.

JFrog ML offers multiple deployment options to suit your business needs, while minimizing maintenance overhead:


JFrog ML Cloud

Deploy your models and features on JFrog ML private and secure cloud environment with a simple ramp up and quicker deployment times. Keep your data encrypted with no added overhead.

With JFrog ML Cloud, you can deploy your first model within minutes!


JFrog ML Hybrid

Keep your data and models stored in your own cloud environment while still taking advantage of a managed solution. JFrog ML provides a CloudFormation template to create a cross-account IAM role, granting JFrog ML the necessary permissions for environment setup and management.

The JFrog ML environment can be installed in 2 ways:

Default (Recommended)

Deploy into a new VPC. The deployment CloudFormation creates all the necessary components to ensure everything works as expected.

Existing VPC

Deploy JFrog ML resources in to an existing VPC environment along with existing resources within the VPC


❗️

We take data privacy seriously.

That's why all data is encrypted by default, ensuring that your machine learning models and data remain private and secure at all times - whether you choose JFrog ML Cloud or Hybrid Deployment.

For more information please refer to our Data & Privacy document.


High Level Architecture

JFrog ML architecture consists of two main components: the Control Plane and the Data Plane.

Qwak high level overview

JFrog ML high level overview

Control Plane

The Control Plane is managed in JFrog ML cloud and handles non-sensitive metadata. It communicates with the Data Plane to delegate sensitive operations such as model and data operations.

Data Plane

The Data Plane hosts the model repository, Feature Store, JFrog ML Inference Lake, and handles the model and data operations (model building, deployment, feature creation, etc).
The Data Plane uses the following services:

  • EKS Cluster to manage JFrog ML services
  • S3 for storing model output data
  • Elasticache to manage the online feature store

What’s Next