Networking a Qumulo Cluster with a Split Networking Configuration

This section explains how to network a cluster with platforms that use a split networking configuration.

Tip
To identify the eth port, run the following command:

for i in /sys/class/net/eth*; \
  do echo $i; \
  cat $i/device/uevent | \
  grep -i pci_slot; \
  done

Prerequisites

Note
Before you create your Qumulo cluster, if your client environment requires Jumbo Frames (9,000 MTU), configure your switch to support a higher MTU.

Your node requires the following resources.

A network switch with the following specifications:
- Ethernet
- Fully non-blocking architecture
- IPv6 capability
Compatible networking cables
A sufficient number of ports for connecting all nodes to the same switch fabric
One static IP for each node, for each defined VLAN

Recommended Configuration

Important
We don’t recommend connecting to a single back-end NIC port because the node becomes unavailable if the single connection fails.

This platform uses a split networking configuration in which different NICs handle back-end and front-end traffic. You can connect the front-end and back-end NICs to the same switch or to different switches. However, for greater reliability, we recommend connecting all four ports on every node: Connect both front-end NIC ports to the front-end switch and both back-end NIC ports to the back-end switch.

We recommend the following configuration for your node.

Your Qumulo front-end MTU configured to match your client environment
One set of redundant switches for the back-end network (9,000 MTU minimum)

Note
You can configure front-end and back-end traffic on the same switch.
One physical connection for each node, for each redundant switch
One Link Aggregation Control Protocol (LACP) port-channel for each network (front-end and back-end) on each node, with the following configuration
- Active mode
- Slow transmit
- Access port or trunk port with a native VLAN
DNS servers
A Network Time Protocol (NTP) server
Firewall protocols or ports allowed for proactive monitoring
Where N is the number of nodes, N-1 floating IP addresses for each node, for each client-facing VLAN

Connecting to Redundant Switches

For redundancy, we recommend connecting your cluster to dual switches. If either switch becomes inoperative, the cluster is still be accessible from the remaining switch.

Front End
- Connect the two front-end NIC ports on your nodes to separate switches.
- The uplinks to the client network must equal the bandwidth from the cluster to the switch.
- The two ports form an LACP port channel by using a multi-chassis link aggregation group.
Back End
- Connect the two back-end NIC ports on your nodes to separate switches.
- Use an appropriate inter-switch link or virtual port channel.
Link Aggregation Control Protocol (LACP)
- For all connection speeds, the default behavior is that of an LACP with 1,500 MTU for the front-end and 9,000 MTU for the back-end interfaces.

Connecting to a Single Switch

You can connect your cluster to a single switch. If this switch becomes inoperative, the entire cluster becomes inaccessible.

Front End
- Connect the two front-end NIC ports to a single switch.
- The uplinks to the client network must equal the bandwidth from the cluster to the switch.
- The two ports form an LACP port channel.
Back End
- Connect the two band-end ports to a single switch.
Link Aggregation Control Protocol (LACP)
- For all connection speeds, the default behavior is that of an LACP with 1,500 MTU for the front-end and 9,000 MTU for the back-end interfaces.

Four-Node Cluster Architecture Diagram

The following is the recommended configuration for a four-node cluster connected to an out-of-band management switch and redundant switches.