PHYSICAL NETWORK DESIGN ON A DCN
The last section of this chapter outlines the basic concepts about the physical network of a DCN. Designing a physical network is the first step in DCN design. You should now have a grasp of the basics of physical networking in the cloud DCN solution and be ready to start to design the physical network.
This section covers the following key points in the physical network design, which can help you design a DCN:
- • Routing protocol selection
- • Server access mode selection
- • Design and principles of border and service leaf nodes
- • Egress network design
Routing Protocol Selection
In most cases, either OSPF or EBGP can be used on an underlay network. OSPF is preferred for most networks. If the scale of the network is large, EBGP is recommended as the underlay network needs to be partitioned into areas, and flexible control of BGP is required.
1. OSPF deployment on the underlay network
When the number of leaf nodes is less than 100, OSPF is recommended on the underlay network. The route planning is as follows: On a single fabric network, OSPF is deployed on all physical switches where spine and leaf nodes are configured. Only OSPF area 0, in which all physical switches are deployed, is planned. OSPF neighbor relationships are established using addresses of Layer 3 routed interfaces to implement connectivity on the underlay network. It is recommended that the network type be P2P, as shown in Figure 5.3.
In a scenario where multiple fabric networks are deployed and constitute a VXLAN domain on the overlay network (that is, a DCN is divided into two fabric networks and is managed through a set of management interfaces), it is recommended that only one OSPF process be deployed on all devices, interconnected devices between
FIGURE 5.3 Recommended OSPF planning for a single fabric network.
FIGURE 5.4 Recommended OSPF planning for multi-fabric deployment (single VXLAN domain).
multiple fabric networks be deployed in OSPF area 0, and Fabricl and Fabric2 be deployed in OSPF areas 1 and 2 respectively. In this case, there is only one OSPF process, and connectivity is achieved on the underlay network, as shown in Figure 5.4.
If multiple fabric networks form two VXLAN domains on the overlay network (that is, two DCNs are managed through two sets of management interfaces, but they need to be interconnected), it is recommended that OSPF be deployed on each fabric network, and interconnected devices between fabric networks exchange routes through BGP.
OSPF is easy to deploy and provides fast convergence for smaller- scale networks; therefore, it is preferred on the underlay network of a small- or medium-sized DON. Most enterprises deploy BGP EVPN on the control plane of the overlay network. When OSPF is selected for the underlay network, OSPF and BGP packets are placed in different queues, and VRFs and routing entries are isolated from each other, which, in turn, keeps fault domains on the underlay and overlay networks isolated.
2. EBGP deployment on the underlay network
When there are more than 100 leaf nodes on a large-scale network, EBGP is recommended on the underlay network as using OSPF would slow down protocol and fault convergence. The route planning is as follows.
FIGURE 5.5 Recommended EBGP planning for a single fabric network.
On a single fabric network, spine nodes are added to the same autonomous system (AS), and each group of leaf nodes is added to an AS. EBGP peer relationships are established between leaf nodes and all spine nodes. As shown in Figure 5.5, EBGP is used to implement Layer 3 interconnection on the entire network.
In a scenario where multiple fabric networks are deployed, the processing method is similar to that for a single fabric network. All spine nodes are deployed in the same AS, as are each group of leaf nodes, and EBGP runs between spine and leaf nodes. Each fabric network is connected to the peer AS through interconnected leaf nodes on the DCN. The interconnected leaf nodes are deployed in an AS and establish peer relationships with spine nodes through EBGP, as shown in Figure 5.6.
The EBGP configuration on the underlay network is complex as groups of leaf nodes, and spine nodes need to be allocated to different ASs, and full-mesh EBGP connections need to be established between spine and leaf nodes. What’s more, TCP-based BGP connections need to be manually specified, which requires heavy configuration workload and the maintenance is complex. However, OSPF processes only need to be enabled on corresponding interfaces when these OSPF processes are being used.
BGP provides an independent routing domain for each area. The fault domain of BGP is smaller than that of OSPF. In addition, BGP provides various route control methods, which increases flexibility
FIGURE 5.6 Recommended EBGP planning for multiple fabric networks.
and facilitates expansion of route control compared with OSPF during network deployment. You can select either of the two protocols to deploy a DCN based on the actual situation.