Network Requirements

General Network Requirements

  • Each host in the deployment must include at least one 10Gb Ethernet card.

  • The network must be configured with either:

    • A DHCP server that supports assigning static IP address by device MAC address.
    • A network that includes a block of static IP addresses reserved out of the DHCP pool.

  • The host names assigned to the servers during installation must be available throughout the life of the deployment.

  • IP addresses assigned to virtual nodes/containers must fall within the floating IP ranges that specified during or after installation.

The other network requirements vary depending on whether your virtual node/container network will use routable or non-routable floating IP address ranges:

Private, Non-routable Virtual Node/Container Addresses

HPE Ezmeral Runtime Enterprise maintains a list of the virtual nodes and ports that users may need to connect to and makes those ports available through one or more Gateway hosts. Each Gateway host maps a range of ports to services running on the containers within the deployment. A user who needs to access a container uses the hostname of the Gateway host and a port number. The hostname can be either the name of a Gateway host or of a physical load balancer.

For example, assume that the deployment has a Gateway worker with the hostname gateway-1.mycompany.com. In this example, the port mappings could appear as follows:

  • virtualnode-1.bdlocal
    • Hive thrift server: gateway-1.mycompany.com:10020
    • MySQL server: gateway-1.mycompany.com:10018
    • Spark master: spark//gateway-1.mycompany.com:10019
    • SSH: gateway-1.mycompany.com:10017
  • virtualnode-2.bdlocal
    • SSH: gateway-1.mycompany.com:10022
  • virtualnode-3.bdlocal
    • SSH: gateway-1.mycompany.com:10024
NOTE
Large deployments might need multiple Gateway nodes. If these nodes share a hostname, then round-robin load balancing will take place among the available Gateway nodes.

Requirements for Using Multiple Subnets

When configuring the container network to use multiple subnets, the following requirements apply:

  • The hosts can be located on-premises, in a public cloud, or both. For example, hosts can reside on multiple racks and/or can be virtual machines residing on cloud-based services (such as AWS, Azure, or GCP).
  • If the deployment includes cloud-based hosts, then the container network must be private and non-routable. The container network is private for Kubernetes deployments.
  • If Platrform HA is configued with a cluster IP address, then the Controller and the Shadow Controller must be on the same subnet.
  • If the Controller and Worker hosts are on different subnets, then the path MTU settings must be the same for both subnets.
  • If the Controller and Worker hosts are on a single subnet and the Gateway hosts are on the different subnet, then:
    • The subnet with the Gateway hosts can use an MTU setting that is lower than or equal to the MTU setting on the other subnet with no further action needed.
    • If the MTU of the Gateway host is larger, then it must be at least 1,000 bytes larger than the MTU setting of the other subnet.

    The subnets used by Gateway hosts can have different path MTU settings, subject to the preceding requirements.

  • If the Controller and Shadow Controller are on one subnet, the Worker hosts on a second subnet, and the Gateway hosts on a third subnet, then each of the hosts must have the same path MTU setting.