Among the different key advanced solutions supported by Dell Enterprise SONiC, the ability to deploy or stand-up a converged or hyperconverged infrastructure in a short amount of time makes a world of difference when trying to gain a business advantage to our customers.
In a converged or hyperconverged infrastructure where all three elements, compute, networking, and storage reside in a single appliance that must work seamlessly while delivering critical applications, the infrastructure design becomes the number one priority.
The following figure shows a standard single site multi-rack VxRail cluster deployment with the Dell Enterprise SONiC NOS in a leaf and spine topology.
The fabric is built using BGP EVPN VxLAN to provide the capability of stretching Layer 2 across an IP fabric. This ability becomes important when the cluster is stretched across racks.
The downstream connections from the leaf switches are Layer 2 dual homed for redundancy and configured as trunks.
From the leaf switches, the uplink connection (L3 or L2) is deployed. The uplink connection is used to integrate with a legacy network and reach DNS and NTP services if using the external IP services option for the VxRail cluster.
The downstream connections speed range between 10 to 100 GbE. The switch interconnect speeds range between 100 to 400 GbE.
Five separate networks with their respective virtual networks are configured, management, vSAN, vMotion, VxRail autodiscovery (3939), and Guest. These networks are stretched across the racks and tagged on the downstream links.
The management network is the only network tagged and trunked on the external connection so it can communicate with the DNS and NTP services.
The jump host connected to a leaf switch is used to connect to the VxRail manager default IP address. The VxRail cluster is deployed through the VxRail manager user interface.
The connection between the jump host and the leaf switch is untagged.
Deployment best practice
Whenever deploying a VxRail environment, the following items must be considered:
- VLAN 3939 must always be present and configured for VxRail nodes to autodiscover each other. This VLAN should be treated as a "Reserved VLAN" inside the fabric and no other data traffic should be used on this VLAN.
- Implement Jumbo frames on the switches.
- Implement trunks and tag the networks on the downstream links connected to the VxRail cluster.
- Implement MC-LAG on the leaf switches.
- The minimum number of leaf interconnect links should be two.
- For Layer 2 uplinks to external switches, ensure that the same link aggregation mode (LACP or static) is configured on both ends of the links.
- Ensure the management VLAN is tagged and trunked on the uplink if Layer 2 uplink is deployed.