Variable Length Subnet Masks
Variable length subnet masks allow an organisation to use different subnet masks within a single network. In other words  to subnet a subnet. This means that IP addresses can be allocated more efficiently and with less waste.
Let's consider the same corporate network again:
The previous subnet design was adequate but caused unacceptable loss of addresses due to the fixed size of each subnet. So I would like to think about the subnet design again. Suppose it were possible to divide the network so that the subnet sizes could vary. Perhaps we could...
 Provide a 64 address subnet for Net A
 Provide a 32 address subnet for Net B
 Provide a 16 address subnet for Net C
 Provide a 4 address subnet for each WAN link
This would be an efficient division of the network address space and would also allow for future expansion. Suppose we were to first divide the network into 4 subnets by borrowing 2 bits from the last octet of the Class C address.
2^{2} = 4 subnets
2^{6}2 = 62 useable hosts addresses
We would have...
Subnet Number 
Subnet Address 
0 
206.12.10.0/26

1 
206.12.10.64/26

2 
206.12.10.128/26

3 
206.12.10.192/26

Net A could go into subnet 0 and Net B could go into subnet 1. This is shown in the diagram below:
This would leave plenty of room for Net A to expand since 30 host addresses would be required immediately, leaving 32 spare for future expansion. However, Net B needs only 20 hosts at the present, leaving 42 spare addresses. This seems rather a lot of spare addresses! Maybe subnet 1 could be used more efficiently?
What if we could divide subnet 1into 2 parts, half for Net B and half for Net C?
This is possible using VLSM. If subnet 1 were to be divided into two then each half would contain 32 total addresses. This would be equivalent of originally dividing up the whole network into eight subnets, I.e.
2^{3} = 8 subnets
2^{5} = 32 addresses of which 30 are useable
Examine the adjusted diagram below and you should see what I mean.
You may wonder why the subnet addresses for Net B and Net C have a new prefix of/27 instead of /26 like Net A. Well, imagine I had divided up the whole network into eight subnets, then this would have meant borrowing 3 bits from the last octet of the Class C address and the so the CIDR prefix would be /27.
Let's get back to the subnet design. We still need subnets for the two WAN links. These are pointtopoint links requiring 2 useable addresses each. What if we could divide the Net C subnet into 2 parts, keeping half for Net C and half for the WAN links? Each half would contain 16 total addresses and one half would certainly be sufficient for Net C.
The other half, containing 16 addresses could be divided even further into four smaller subnets, each containing 4 addresses each. Two of these small subnets could be used for the WAN links and the other two subnets could be held in reserve. Examine the adjusted diagram below and you should see what I mean.
Again, you may wonder why the prefix for Net C has changed from /27 to /28. Well, imagine I had divided up the whole network into 16 subnets, each containing 16 addresses, then this would have meant borrowing 4 bits from the last octet of the Class C address and the so the CIDR prefix would be /28.
Why has the subnet prefix for each WAN link changed to /30? Each WAN link contains 4 addresses. This is as if I had divided up the whole network into 64 subnets, each containing 4 addresses. This would have meant borrowing 6 bits from the last octet of the Class C address and the so the CIDR prefix would be /30. I.e.
2^{6} = 64 subnets
2^{2} = 4 addresses of which 2 are useable
Here's how the subnets look in more tabular form...
This division of the total address space into subnets of variable size leads to a more efficient and flexible subnet design solution. For this particular network, there are a few different ways to divide the network using VLSM and I have illustrated just one solution. The use of VLSM has maximized the number of addresses and eliminated the waste from two pointtopoint WAN links using a /27 subnet each.
Although there may be a variety of VLSM subnetting solutions for a particular network, the challenge is to choose the best division and subdivisioning solution to meet an organisation's present and future needs.
~Now try the activity~
Activity A 
 Using VLSM, divide the network as efficiently as possible to meet the following requirements:
 A subnet for LAN A with at most 62 useable addresses
 A subnet for LAN B with at most 30 useable addresses
 A separate subnet for the WAN link
You can check your answers here

