Network configuration

The Cloud Pak for Data System networking is mostly dynamic. There is a fixed Control Plane which consists of 3 nodes. These 3 nodes provide DHCP and DNS services to the rest of the cluster. The Worker Nodes, Virtual Machines (control and worker), and management devices all use dynamically assigned addresses.

There are 2 different networks:
Management network
  • Provides monitoring and similar functionality
  • Management Network in a Base System is based on Mellanox AS4610 switch with 1G UTP links to each node.
  • Management switch is managed by Platform Manager
  • No redundancy
  • Failsafe management access is provided to the first 3 (control plane) nodes in the system

Certain customer traffic patterns are forced out the customer network via policy based routing. These traffic patterns include:

  • upstream DNS
  • upstream LDAP
  • upstream NTP (time)
  • upstream SMTP
  • Callhome
Figure 1. Management switch connections
Image showing management switch connections
Fabric network
  • Provides a higher speed and bandwidth application layer communications
  • Fabric Network in a Base System is based on Mellanox SN2410 switch with 25GBASE-SR links to each node and the house.
  • Data Fabric Network is active-active 25Gb connections to the Innova-2
  • Fabric switch is managed by Platform Manager
  • There are redundant links to a single card
Figure 2. Fabric switch connections
Image showing fabric switch connections

The customer access is via the fabric network, and is routed through the OpenShift console. All activity which occurs on the cluster is accessible through the console.

There are a minimum of 5 customer IP addresses in a standard system:
  • Node1 management
  • Node2 management
  • Node3 management
  • Management Float
  • Application Float
All activity should be completed via the Application Float. Each node can also be referenced by its specific management IP in the case of maintenance.

The nodes are named after their enclosure and index. The first four nodes are e1n1, e1n2, e1n3, and e1n4. The second enclosure contains the next four nodes: e2n1, e2n2, e2n3, and e2n4. The index portion of the node names are always relative to their enclosure.

Each of the bare metal nodes also has a virtual machine. The virtual machines are for control or application work depending on their host. The VM names are based on their host (eEnN), slot number (an index, 1-N), and their role (control or worker). So the first VM is e1n1-1-control.

Each node (bare metal and virtual) has an interface in the mbond and fbond domains. The Fully Qualified Domain Name is either e1n1.fbond or e1n1.mbond depending on which interface you are using.

High availability

You can expand the base system with additional management and fabric switches for high availability, and add more expansion units as described in Expanding Cloud Pak for Data System.

If Cloud Pak for Data System is configured with HA switches, bonding must be used. Single ports are not supported. Link Aggregation Control Protocol (LACP) 802.3ad is the only supported bonding mode.