This section discusses the similarities and differences between CDP Private Cloud networking requirements and those provided for Cloudera Manager Virtual Private Clusters (CM VPC).
At a high level, the requirements are similar, in that the model of deployment in both cases are similar -- a base cluster which houses HDFS storage and remote compute-only clusters which would be able to read from and write to the base HDFS cluster.
In CDP private cloud, the network bandwidth requirements are less stringent than those of the Cloudera Manager Virtual Private Cluster (VPC) because of data caching technology being introduced at the compute layer, which is not available in VPCs.
This implies that while the initial load of data from the remote storage would require significant bandwidth between the compute and storage clusters, subject to quantity of data being ingested; subsequently, the network bandwidth requirements should be significantly lower.
As already mentioned, a minimum of 1Gbps (recommended 10Gbps) guaranteed bandwidth is required between each OpenShift worker node and each CDP Data Center DataNode. To test the worst case bandwidth, the network should be fully stressed with all OpenShift nodes trying to read or write from the Data Center nodes at the same time.
The recommended network architecture is Spine-Leaf with no more than a 4:1 oversubscription between the spine and leaf switches.
For more information including recommendations on network topology, refer to Networking Considerations for Virtual Private Clusters | 6.3.x.