Skip to main content

Deploy xCloud on your VPC

xCloud offers you the flexibility to harness the power of your own Virtual Private Cloud (VPC) for enhanced control and data privacy. Please, don't hesitate to reach out to us at xcloud-support@stochastic.ai if you are interested on this feature.

Here's what you need to know:

1. Advantages of Using Your Own VPC

The advantages of using xCloud on your own VPC are:

  • Utilize your own hardware: leverage your existing infrastructure to maximize resource utilization.
  • Keep your data and models within your cloud environment: ensure data security and maintain control over your valuable assets.

2. High-Level Architecture

xCloud consists of two main components:

  • Control Plane: receives user requests for creating, updating, or deleting a resource.
  • Compute Plane: it runs within your own VPC. This component enables you to use your hardware while ensuring your data remains within your VPC.

3. Data privacy

The control plane will store the following information:

  • Metadata: includes all information provided during resource creation, such as name, creator, code location, Docker image, commands to be executed within the Docker image, etc.
  • Logs of running containers.

It won’t store:

  • Code, data or model used to run the resources.
  • Secrets attached when creating a resource.

4. Launch a resource on your own VPC

Once we have setup and linked your cloud environment with our platform, you will be able to create resources (notebooks, executions and deployments) in your cloud environment. For that you will have to specify the link_name and the flag on_premise to True. Below you can see some examples.

For notebooks:

notebook = Notebook(
    notebook_name=notebook_name,
    container_specs=container_specs,
    link_name=link_name,
    on_premise=True
)

For deployments:

deployment = Deployment(
    deployment_name=deployment_name,
    model_specs=model_specs,
    container_specs=container_specs,
    deployment_specs=deployment_specs,
    link_name=link_name,
    on_premise=True
)

For executions (jobs):

job = ExecutionJob(
    job_name=job_name,
    credentials=creds,
    code_specs=code_specs,
    container_specs=[container_specs],
    link_name=link_name,
    on_premise=True
)