Recently I came across MAAS and I tried it in combination with juju and it’s great! So easy to manage and automate things, ridiculous! But now I’m wondering how an appropriate network topology for large scale deployments would look like?
Let’s imagine you’ve got the following scenario, that should be handled by using MAAS:
- building a private cloud
- 500 server racks
- 40 machines each
- each rack has it’s dedicated rack controller and ToR-switch(es)
First I started using a
10.0.0.0/8 network to get all those addresses fitting and have some scale options for future needs. But in a different discussion, I noticed using such a big network gets inperformant very quickly. For those purpose, each rack should have its own dedicated subnet, let’s say
10.0.3.0/24 and so on, which can be interconnected defining a space.
What I do not understand is, in which network are all those rack controllers placed? Within each subnet or the underlying
10.0.0.0/24 network? Another question arised: how to adjust those architecture to fit more racks let’s say 1000 and more into the same maas deployment? Using a
/24 CIDR in combination with a
10.x.x.x address space would allow 254 * 2 subnets, wouldn’t it?
Many thanks in advice, feel free to ask questions if something Isn’t understandable