Differences Between Public and Private Cloud
There are some differences in Cloudera Machine Learning functionality between Public and Private Cloud.
Feature | Public Cloud | Private Cloud 1.5.x |
---|---|---|
CML application control plane (infrastructure containers and workload containers) | Control plane is hosted on public cloud servers. | Control plane is hosted on customer’s cluster. |
Storage - CML internal state data (database, images, logs) | EBS on AWS, Azure Disks on Azure. | Software Defined Storage System, such as Ceph or Portworx. |
Storage - User project files | EFS on AWS, external NFS on Azure. | Internal NFS storage is recommended. |
Autoscaling | CPU/GPU nodes scale up and down as needed. | Autoscaling concept is different; Private Cloud shares a pooled set of resources among workloads. |
Logging | Per-workspace diagnostic bundles can be downloaded from the workspace. | Diagnostic bundles are not supported at Workspace level, but can be downloaded from the control plane at the cluster level. |
Monitoring dashboards | Provides four dashboards. | Provides two dashboards, for K8s Container and K8s Cluster. |
NFS support | AWS uses EFS; Azure requires external NFS. | Internal NFS is recommended, external NFS is supported. |
TLS support | TLS access to workspaces is supported. | TLS access is supported, but requires manual setup of certificate and other steps. |
Hadoop Authentication | Uses FreeIPA | User needs to provide credentials to communicate with the CDP Private Base cluster. |
Remote Access | Available from each workspace. | Not available in the workspace. Instead, the environment's kubeconfig
file may be downloaded from Environments using the Download
Kubernetes configuration action for the specified environment. |
Roles | MLAdmin, MLUser | The corresponding roles are: EnvironmentAdmin, EnvironmentUser |