On-premise Deployment Options

Scaleout Edge is designed to scale from a single laptop to a multi-region cloud cluster. To support this flexibility, we offer two primary deployment architectures:

  • Starter (Docker Compose): For R&D, proof-of-concept, and small-scale pilots.

  • Production (Kubernetes): For mission-critical, high-availability enterprise deployments.

1. Starter: Docker Compose

The Starter deployment runs the entire Control Plane (UI, Controller, Combiner, Database) on a single machine using Docker Compose.

Use Cases

  • R&D & Prototyping: Rapidly iterate on model architectures and client logic locally.

  • Small Fleets: Manage up to ~50 edge nodes for a pilot study.

  • Air-Gapped Demos: Run a complete, self-contained network without internet access.

Architecture

In this mode, all services run as containers on a shared Docker network:

  • Postgres: Stores metadata, users, and audit logs.

  • MinIO: Acts as a local S3-compatible object store for models and compute packages.

  • Controller & Combiner: Run as lightweight services behind a reverse proxy that terminates TLS for both HTTPS and gRPC and forwards plain traffic to the backend.

This is the default layout. Ports, TLS and the proxy in front can be changed. See Self-managed Deployment for the options.

2. Production: Kubernetes (K8s)

For enterprise scale, high availability, and strict security requirements, Scaleout Edge should be deployed on Kubernetes. We provide official Helm Charts to automate this process.

Use Cases

  • Large Scale: Orchestrating thousands or millions of edge nodes.

  • High Availability: Redundant Controllers and Combiners to ensure zero downtime.

  • Enterprise Integration: Connecting to external identity providers (OIDC/LDAP), managed databases (RDS/CloudSQL), and observability stacks (Datadog/Prometheus).

Architecture Changes

Moving to production introduces several architectural enhancements:

Component

Starter (Docker)

Production (K8s)

Orchestration

docker-compose.yml

Helm Charts

Storage

Local MinIO Container

Managed Object Storage (AWS S3, Azure Blob, GCS)

Database

Local Postgres Container

Managed Database (Postgres)

Ingress

Bundled NGINX reverse proxy with TLS termination

K8s Ingress Controller (NGINX/Traefik) with TLS termination

Scaling

Single Instance

Horizontal Pod Autoscaling (HPA)

Deployment Workflow

Deploying to a K8s cluster (e.g., EKS, AKS, GKE) typically involves:

  1. Configure values.yaml: Define your domain, external database credentials, and resource limits.

  2. Install via Helm:

helm repo add scaleout https://charts.scaleoutsystems.com
helm install scaleout-edge scaleout/scaleout-edge -f values.yaml

  1. Scale: As your fleet grows, simply increase the replicaCount of the services (or use Horizontal Pod Autoscaling) to handle more concurrent connections.