Home > Networking Solutions > Other Networking Solutions > General Networking Guides > Guides > Dell EMC Networking with Isilon Front-End Deployment and Best Practices > Scaling
An example of scaling this solution in a two-tier leaf-spine is a configuration of up to 16 racks. The Dell EMC Z9100-ON has 32x 40/100 GbE interfaces that can support 16 leaf pairs using VLT. This provides one rack that contains WAN-edge connectivity and 15 racks for servers and storage nodes. Each rack of the compute/storage rack houses a combination of up to 19 PowerEdge R730s or Isilon X210s.
In this particular example, each R730 has four 10 GbE uplinks, and each Isilon node has four 10 GbE uplinks with 19 servers/nodes per rack. Additionally, the example architecture has four spine switches to minimize oversubscription.
Table 2. Connections for 16 racks with 4 spine switches
|
Server/storage interfaces |
Server connections to leaf switches |
Leaf connections to spine switches per rack |
Total connections for leaf switches to four spine switches |
Connections |
4 |
19 x 4 = 76 |
4 per leaf switch, 2 leaf switches per rack = 8 links |
16 racks x 8 = 128 |
Speed of ports |
10 GbE |
10 GbE |
40 GbE |
40 GbE |
Total theoretical available bandwidth |
4 x 10 = 40 GbE |
76 x 10 = 760 GbE |
8 x 40 GbE per rack = 320 GbE |
16 x 320 GbE = 5120 GbE |
This example provides for an oversubscription rate of 2.375:1 using 40 GbE spine connectivity.
Scaling beyond 16 racks would require a three-tier leaf-spine network. The proof-of-concept scaling shown in the following figure allows four 16-rack pods, connected using an additional spine layer, to scale in excess of 1,000 nodes with the same oversubscription ratio. This scenario requires reducing the number of racks available per pod to accommodate the uplinks required to connect to the super spine layer.
It is important to understand the port-density of switches used and their feature sets’ impact on the number of available ports. This directly influences the number of switches necessary for proper scaling.
Figure 11. Scale out of the existing networking topology