skip to main content
Technical Operations > ICE OS > Kubernetes > Kubernetes Deployment Models > Choosing a deployment model
Choosing a deployment model
The two tables below illustrate the benefits and drawbacks of choosing a deployment model and cluster architecture.
Deployment Table
Deployment
Benefits
Drawbacks
Cloud hosted
Cost and complexity of building, administrating and maintaining a Kubernetes cluster is delegated to the cloud provider. Installation can be completed in minutes.
Assumes ICE clients are able to reach the Internet. Some IT departments may perceive solution as more expensive than utilizing their own hardware.
On premises
System can be deployed on remote or highly secured ('air gapped') networks that are not connected to the Internet.
Substantially increases administrative and maintenance burden on the administrator. Administrator becomes responsible for creating the Kubernetes cluster and assuring those machines comply with the organization’s security and IA policies.
Cluster Table
Cluster
Benefits
Drawbacks
Single node cluster (ICE OS), single site
Simplest, least resource-intensive deployment model. Ideal for small scale systems, labs, demo environments and mobile systems (i.e., vehicle mounted).
Provides no hardware resiliency. Should the physical machine running ICE fail, the system will suffer a full outage.
Multi node cluster, single site
Allows organizations to increase system scale over time by adding nodes to the cluster. Provides full resiliency between nodes in the cluster: Any node can fail with little or no user-visible impact to operation.
Additional complexity and compute resources: A minimum of three physical servers are required to provide hardware level resiliency.
Two clusters, geographic redundancy
Provides extra resiliency when an entire data center goes offline. ICE clients automatically reconnect to the surviving data center and continue operating normally.
Most complex and resource intensive deployment; requires double the compute resources of a single-cluster setup and demands a high-speed link between clusters. Not all functions of Instant Connect are fully resilient, even when using this deployment model.