VXLAN Protocol introduction

Question

[RFC 7348] about VXLAN in the introduction says:

Data centers are often required to host multiple tenants, each with their own isolated network domain. Since it is not economical to realize this with dedicated infrastructure, network administrators opt to implement isolation over a shared network.

I imagine a rack with multiple servers, each one running multiple virtual machines(VMs). All VMs owned by a single server/tenant, compose its own isolated network domain? What is not economical to realize?

Answer 1

There are a number of difficulties with physically separate networks.

1. Resources for a tenant may be spread out through the datacenter due to gradual growth over time or a desire not to "put all your eggs in one basket". Running long dedicated cables for each client gets expensive.
2. Resources are likely to be reallocated over time. Physical re-patching when resources are reallocated gets expensive and time consuming.
3. One VM host may host VMs for many different clients. Fitting a seperate NIC to the VM host for each client would get prohibitively expensive.
4. There is a certain minimum cost for a router or switch. Having switches with only a handful of ports used gets expensive.

So customers want logically separate Ethernet networks but physically separate networks end up being prohibitively expensive. So we need to build logically separate Ethernet networks over the top of a common physical network.

VLANs provide a method for doing this but they have scalability problems. VXLAN aims to provide a more scalable solution (at the cost of substantially higher encapsulation overhead).

Answer 2

If you read more you can find this in 3.2. Multi-tenant Environments

Cloud computing involves on-demand elastic provisioning of resources
for multi-tenant environments. The most common example of cloud
computing is the public cloud, where a cloud service provider offers
these elastic services to multiple customers/tenants over the same
physical infrastructure.

The vms for different users/tenants might be placed in a single server.

Answer 3

There are various reasons when servers that should be in a single segment are geographically remote, including:

• hosted servers are often not deployed at the same time; later deployment usually means another rack, possibly even another data center
• servers are put into different fire zones or data centers on purpose for availability/fail-over reasons
• hosted servers or services need to be combined with on-premise installations

When servers are physically remote to each other there might still be good reasons to arrange them very tightly logically. Without virtual network connections this means running separate network links for each zone. This increases cost and can be prohibitively expensive.